AMD and AI infrastructure company Cerebras have unveiled a joint system integrating AI infrastructure from both companies.

The joint solution, which was announced at AMD Advancing AI 2026, features AMD Helios – the star product of the show – and the Cerebras Wafer-Scale Engine.

According to Cerebras and AMD, it's intended to solve the issue of inference workloads having variable requirements when it comes to latency, throughput, token capacity, cost, and scale.

For example, agents and copilots need faster response times, whereas high-volume workloads prioritize maximum code generation, the organizations said.

By unifying Helios and Wafer-Scale Engine, the system optimizes the two stages of workflow independently: Helios is tuned to process prompts and large context windows, providing ultra-high throughput.

Wafer-Scale Engine, meanwhile, accelerates memory and bandwidth-intensive token generation with ultra-low latency.

“The demand for ultra-fast inference is growing at an unprecedented pace,” said Cerebras CEO and co-founder Andrew Feldman. “Partnering with AMD gives us an incredible opportunity to bring that performance to even more customers.”

AMD CEO Lisa Su added: “AI inference is becoming one of the largest infrastructure opportunities in AI and its growing diversity requires a more flexible approach.”

“Together with Cerebras, we are extending that leadership into the most latency-sensitive applications and creating a powerful new platform for real-time agentic AI.”

The joint solution will be available in the second half of 2026, initially through Cerebras Cloud.

The announcement comes alongside a number of other Helios partner announcements, including new commitments from Microsoft and Anthropic to roll out the system to their infrastructure.

FOLLOW US ON SOCIAL MEDIA

AMD has launched its latest Instinct GPUs, the MI455X, in tandem with bringing its Helios AI architecture into production at AMD Advancing AI 2026.

Helios, which was announced at last year’s annual conference, is the company’s flagship hardware product, integrating Epyc CPUs, Instinct GPUs, Pensando networking, and Rocm software stack.

“Helios is really the system that scales to be an AI factory,” said Andrew Dieckman, CVP and GM of data center and GPU at AMD. “We set out to build the world’s highest-performing rack and we believe we have delivered the world’s highest performing rack.”

According to AMD, Helios has 15% more AI compute, 50% more memory, and 50% more scale-out bandwidth than Nvidia’s competitor offering, Vera Rubin NVL72.

At the heart of Helios – the “engine” in Dieckman’s words – is the new AMD Instinct MI455X CPU. The MI400 series features increased HBM4 capacity and bandwidth, expanded AI formats with higher throughput, and standard-based rack-scale networking.

In Helios, the MI455X acts as the accelerator, delivering 40 petaflops of peak FB4 compute, with 423GB of HBM core and a peak memory bandwidth of 23.3 terabytes per second of memory per GPU.

In addition to the new MI455X GPU, Helios also features 6th Gen AMD Epyc server CPUs with 96 high-frequency cores and 1.6TB/s of memory bandwidth, as well as AMD Pensando and UALoE networking fabric.

This can scale up to 260TB/s bandwidth and host 72 GPUs on one domain.

AMD showcases Helios partners

With Helios already in production, AMD was keen to show off some of its partners that are deploying – or pledged to deploy – the system. These include OpenAI, Meta, Oracle, Microsoft, Dell Technologies, HPE, Lenovo, Bull, Cisco, Tensorwave, and more.

“Helios is a big step forward for AMD because we’re moving to a full rack-scale architecture where we provide all the components in a really co-designed engine and co-engineered rack, so that [customers] can get the full benefit out of the complete AMD end-to-end solution,” said Dieckman.

“We continue to be very focused on driving improved economics for our customers – making sure that we deliver the most tokens per dollar within the market.”

FOLLOW US ON SOCIAL MEDIA

Tokenomics has been a theme of many tech conferences this year and AMD has joined in the fun at this year’s Advancing AI conference.

At the heart of the tokenomics conversation is the move from chatbots and simple inference towards agentic AI. This is a real trend as noted by Digital Ocean and McKinsey and one that OEMs have locked onto hard.

While in previous years enterprise hardware vendors would wax lyrical about the importance of training LLMs on your own data (and how their hardware can help), if you listen for that chatter now you will hear nothing but crickets.

AI agents offer a range of benefits for enterprises – according to Digital Ocean’s research, 53% of the over 1,000 IT decision makers it spoke to saw productivity and time saving gains, while 44% said it had created new business opportunities.

There are also drawbacks, however, and one of the most apparent is the increased use of tokens, which leads to higher bills.

Whereas an individual using prompts to interact with an AI chatbot may use a few hundred tokens, agents can use up thousands.

As McKinsey explains: “Agentic tasks can consume roughly 1,000 times more tokens than code reasoning (single-turn problem-solving without tool interaction) or chat tasks (multi-turn dialogue about a coding problem).”

“The expensive part is not the first answer generated but the checking, repairing, and reverifying that follows,” the report continues. “About 60% of an agentic task’s costs, in fact, are tied to refining answers.”

This fact, and companies baulking at the related costs, hasn’t gone unnoticed among enterprise IT vendors, which are proposing their on-premises services as a way round this.

For AMD, this is a tricky path to tread, given it counts both traditional enterprise vendors like Dell, HPE, Lenovo, and Supermicro and public cloud-based services from OpenAI and Anthropic among its customers.

AMD’s performance focus

Michael Nordquist, CVP of product marketing for AMD’s computing and graphics group, diplomatically said that when it comes to considering the impact of tokenomics “it’s not all one way or all the other, it’s going to be hybrid. For some things local models work great, for some things you’re going to want that frontier model available for agent behaviour.”

From a practical perspective, AMD’s answer to this quandary has been to focus on how its own chips can (theoretically) reduce token-related costs, without taking sides on the public/on-premises debate.

“The theme [at Advancing AI] is really performance is the way that we can deliver the best economic value to our customers,” said Andrew Dieckman, CVP and GM for data center GPU at AMD.

“It’s the strongest dial that we have on the dashboard, so to speak,” Dieckman said, “so we’re very focused on that and then making sure that conveys to favorable economics for our customers because the cost of tokens as the world moves to inference is so important.”

Part of that focus is the company’s MI455X GPUs, announced at Advancing AI 2026 during AMD CEO Lisa Su’s keynote speech.

According to AMD, the MI455X has up to 18x lower token cost than its predecessor, the MI355X, while maintaining up to 34x higher token throughput.

“That delivers, really, the ability to serve more users with faster response times and significantly better economics,” Dieckman said.

FOLLOW US ON SOCIAL MEDIA

Microsoft has announced it will deploy AMD’s Helios rack-scale solution as it targets data processing and AI inference gains for Azure customers.

The collaboration will see three new AMD EPYC CPU-powered virtual machine (VM) offerings made available through Azure. These include:

• HDv2 VMs
• HCx2 VMs
• ND MI455X v7 VMs

The first of these, Azure HDv2, is designed for large-scale data processing and to “empower massive agentic workload adoption”.

Featuring 500 sixth-generation EPYC CPU cores, HDv2 VMs boast 4TB of RAM, 32TB of local NVMe storage, and 500Gbps Azure Boost networking capabilities.

Azure HCx2 VMs, meanwhile, are designed primarily for silicon design and technical computing tasks, such as scientific simulations and engineering analysis.

This series builds on the design of Azure HX virtual machines, which the company launched in partnership with AMD in 2023.

HXv2 VMs will include 176 AMD sixth-gen EPYC CPU cores. These offer “significantly increased” per-VM and per-core performance and will also include 800GB InfinBand for “large-scale MPI-based simulations”.

Elsewhere, the ND MI455X v7 specifically targets “production-scale” AI inference improvements for Azure customers.

“ND MI455X v7 is designed for the reasoning, search and agentic workloads behind modern AI services,” Microsoft said in a blog post.

“Powered by the AMD Helios rackscale solution, it expands Azure’s infrastructure options for large-scale inference and is designed to deliver strong performance and efficiency for demanding AI workloads.”

The move makes Microsoft the latest in a string of big tech customers flocking to the Helios system. AMD chief executive Lisa Su said the deal builds on long-standing ties between the duo.

“AMD and Microsoft have spent years building high-performance infrastructure together, and today we're extending that partnership across the full stack of AMD AI solutions on Azure,” said AMD CEO Lisa Su.

“Microsoft's new AMD deployments mark an important milestone as we deliver leadership compute solutions to Azure customers and scale the next generation of AI infrastructure together.”

Microsoft targets “scale and choice” for customers

Helios is AMD’s first rack-scale system designed specifically for frontier model training and large-scale AI inference. The system combines AMD Instinct MI455X GPUs, AMD EPYC “Venice” CPUs, Pensando networking, and ROCm software..

Microsoft CEO Satya Nadella said the collaboration marks a significant expansion of Azure infrastructure capabilities.

“Customers are looking for AI infrastructure that is optimized for a wide range of workloads, from training and inference to data preparation, search, and reinforcement learning," he said.

"Through our collaboration with AMD, we are expanding the Azure infrastructure portfolio with AMD Helios to give customers the performance, scale and choice they need to build and run the next generation of AI applications.”

Helios roll-out pending

AMD unveiled the Helios rack-scale solution at its 2025 Advancing AI conference. The solution offers an alternative to – and competes with – Nvidia’s Vera Rubin and Grace Blackwell systems.

At the time, Su described the launch of Helios as a “game changer” that’s “purpose built for the most demanding AI workloads”.

A host of major tech providers, including Meta, OpenAI, HPE, and Oracle have since confirmed they’ll deploy the solution in data centers.

The chipmaker will begin shipping Helios to customers in the second half of 2026.

Although exact timelines haven’t been revealed, the collaboration with Microsoft comes just days ahead of AMD’s 2026 Advancing AI conference, in which customers can likely expect real-world examples of the system in action.

ITPro will be live on the ground in San Francisco for AMD’s annual conference this week.

Stay up-to-date with all the news, views, and announcements from the event by following our socials (below) and subscribing to our daily newsletter.

FOLLOW US ON SOCIAL MEDIA

AMD Advancing AI 2026 takes place this week at the Moscone Center in San Francisco, with the two-day event largely focusing on the second day, when we will hear from AMD’s CEO Lisa Su.

One notable thing, for me, is a sense that in many ways AMD and the wider chip market is in largely the same position as it was this time last year. Nvidia is still the talk of the town when it comes to AI infrastructure, with CEO Jensen Huang once again guesting at other hardware vendors’ flagship events.

AMD still has HPC pedigree (more on that below) but doesn’t have the cachet Nvidia does, while Intel trails further behind.

Despite this, there are still predictions to be made, so here are three things I expect to see at AMD Advancing AI 2026 and one I don’t.

Helios

AMD Helios, the company’s fully-integrated, rack-scale architecture solution, was unveiled at Advancing AI 2025 but we are yet to see any deployed (at least publicly).

This is despite HPE saying it would have a cluster available in its cloud in the second half of 2026 (for those counting, that’s now) and Oracle stating it will build its AI superclusters on Helios.

What better time to talk about Helios being out in the market than at Advancing AI?

My feeling is it will build on last year’s announcement by wheeling out some real-world examples of the hardware in use, whether as part of the HPE or Oracle partnerships or something else.

Altair MI450 chips

Annual company events in the tech industry are for product launches as much as anything else. Having already made several announcements about its gaming portfolio at Computex 2026 in May, we can probably discount major announcements around Ryzen or Radeon – although there may be a recap.

Instead, one of the flagship announcements could be the company’s highly anticipated next-gen GPU series the MI450 – code named Altair.

Oracle already announced in October 2025 that it would be putting MI450 into OCI once it’s available and rumour has it the company has already inked a deal to sell these chips to Anthropic. Speaking of which…

AI partnerships

AMD – perhaps more than other hardware companies – makes a lot of its relationship with AI businesses.

This includes the likes of Anthropic and OpenAI, with the latter’s CEO, Sam Altman, appearing on stage at Advancing AI last year, as well as Hugging Face, PyTorch, and Llama.

I also expect to hear more about what role AMD expects to play in the brave new world of agentic AI

Supercomputing and HPC

While we might hear about high performance computers (HPC) – for example Alice Recoque, which is supposed to start onboarding this year – there may be little conversation around supercomputing more generally.

This is because while an AMD machine has been ranked number one in the Top500 list of most powerful supercomputers consistently since 2022 and held the top two places since November 2024, it was overtaken by a Chinese supercomputer in the June 2026 rankings.

AMD Advancing AI 2026 takes place from 22-23 July, with CEO Lisa Su’s keynote happening on the morning of the second day. I will be on the ground reporting on all the news coming out of the event and providing the usual dose of analysis you have come to expect.

Follow here for all previous years’ coverage and everything from this year as it comes in.

AMD has announced plans to invest £2 billion in the UK as the chip maker aims to accelerate AI innovation and research, and expand access to compute resources.

The five-year plan, announced at London Tech Week, will see the firm form new strategic partnerships with Imperial College London and Oriole Networks, along with support for the University of Cambridge's Zenith AI supercomputer and Sunrise fusion AI system.

“This investment is a major vote of confidence in Britain’s place as a global AI superpower. We’ve got the talent, the world-class universities and the ambition to lead, and partnerships like this help turn that potential into real progress,” said chancellor Rachel Reeves.

"It will drive more cutting-edge research here in the UK, open up opportunities for people to build the skills they need for the jobs of the future, and speed up breakthroughs that can improve people’s lives and grow our economy.”

The collaboration with Imperial College London is aimed at advancing computational science and supporting research that relies on large-scale computing resources, including healthcare innovation and climate modeling.

The two also plan to explore opportunities to optimize AI models, scientific workflows and data-intensive applications on AMD compute platforms and AMD ROCm open software.

“The United Kingdom has the talent, research excellence and ambition to help lead the next era of AI,” said AMD chief executive Lisa Su.

“AMD is proud to deepen our commitment to the UK and work with partners across government, academia and industry to expand access to the compute infrastructure needed to advance sovereign AI, accelerate discovery and drive long-term economic growth.”

AMD pens Oriole Networks deal

AMD has also announced a collaboration with Oriole Networks in support of the UK's Advanced Research and Invention Agency (ARIA) Scaling Inference Lab.

This is a national initiative designed to address critical AI infrastructure bottlenecks and support what's expected to be the world's first large-scale AI system powered by a pure photonic network.

AMD and Dell Technologies, meanwhile, are working with the University of Cambridge on expanding national AI infrastructure. This includes plans to develop the new Zenith AI supercomputer and the Sunrise fusion AI system developed in collaboration with the United Kingdom Atomic Energy Authority (UKAEA).

Zenith is a new UK AI-for-science platform, funded by the Department for Science, Innovation and Technology (DSIT) and UK Research and Innovation (UKRI), while Sunrise is a second AI supercomputer being built now, funded by the Department for Energy Security and Net Zero (DESNZ), owned by UKAEA, and operated by the University of Cambridge.

Together, they will support a broad range of AI-for-science applications including healthcare research, climate modelling, materials science, engineering simulation, fusion research, and scientific AI model development.

The AMD investment comes alongside the launch of a new £1.1 billion government plan to boost the UK's chip industry and attract more AI funding.

“This investment reflects the strength of Britain’s talent, research and ambition in AI - but also the infrastructure we are putting in place to match it,” said technology secretary Liz Kendall.

"With world-class chip designers, leading universities, and partners such as AMD choosing to invest here, we are building the compute capability needed to power innovation, drive growth, create jobs, and ensure the most advanced AI technologies are developed in the UK."

FOLLOW US ON SOCIAL MEDIA

AMD and Samsung Electronics have announced plans to team up on “next-generation” chip development for AI.

The agreement between the duo will see Samsung supply high-bandwidth memory (HBM4) for AMD’s Instinct MI455X AI accelerators, the companies said in a statement.

AMD chief executive Lisa Su said the agreement aims to drive collaboration amidst a sharpened focus on AI infrastructure development.

“We are thrilled to expand our work with Samsung, bringing together their leadership in advanced memory with our Instinct GPUs, EPYC CPUs and rack-scale platforms,” she said.

The partnership will also ramp up integration “across the full computing stack, from silicon to system to rack”, Su added.

Young Hyun Jun, vice chairman and CEO of Samsung Electronics said the company is “uniquely positioned to deliver unrivaled turnkey capabilities that support AMD’s evolving AI roadmap”.

Closer ties

The new agreement builds on long-standing ties between the two companies, with Samsung already providing AMD with HBM3E chips used to power AMD’s MI350X and MI355X accelerators.

Samsung’s HBM4 is built using its 6th-generation 10-nanometer (nm) DRAM process (1c) and a 4nm logic base die, boasting processing speeds of up to 13 gigabits per second, as well as a maximum of 3.3 terabytes per second bandwidth.

AMD said this will be crucial in underpinning the Instinct MI455X GPU range, which is designed specifically for use in high-performance systems that handle AI model training and inference.

MI455X GPUs form a key component of AMD’s Helios open rack-scale architecture, a platform designed for AI workloads, which is due to launch in late 2026.

The partnership will also see the pair collaborate on development of optimized DDR5 memory for AMD’s sixth-gen EPYC CPUs, also for use in the Helios architecture.

Crunch time for chipmakers

As part of the agreement, Samsung said the two companies will also “discuss opportunities” for a foundry partnership, with the South Korean firm providing services for future AMD products.

This aspect of the agreement comes as Nvidia fosters closer ties with the company.

Speaking at Nvidia GTC 2026 this week, CEO Jensen Huang noted Samsung is manufacturing the Groq 3 LPU, which is set to be integrated within the firm’s Vera Rubin architecture.

The deal also comes as the tech industry contends with acute memory and storage shortages. Samsung previously warned that the memory shortage could continue into 2027, while SK Hynix also issued a similar warning.

FOLLOW US ON SOCIAL MEDIA

The UK government has announced plans to invest £45 million to build a supercomputer designed to support nuclear fusion research.

The 1.4 MW facility, named Sunrise, will be located at the UK’s first AI Growth Zone at the UK Atomic Energy Authority’s (UKAEA) Culham Campus in Oxfordshire.

It will deliver up to 6.76 Exaflops of AI-accelerated modeling, with the aim of enabling high-fidelity simulations and creating digital twins for complex systems.

"UKAEA is taking lessons from the Apollo programme: we learn fastest when we can test, iterate, and improve safely in the virtual world before we commit to our real-world mission," said Rob Akers, UKAEA’s director for computing programmes.

"Sunrise will bring that capability to fusion by combining high-fidelity simulation with physics-informed AI to develop predictive digital twins that reduce the cost, risk and time of learning that would otherwise require expensive and time-consuming physical testing."

What will Sunrise be used for?

The aim is to tackle key fusion energy challenges in areas such as plasma turbulence, materials development, and tritium fuel breeding, while delivering additional benefits to other clean energy technologies and the UK’s broader net zero ambitions.

The supercomputer will be used within several UK fusion initiatives, including the Lithium Breeding Tritium Innovation (LIBRTI) program, which is developing tritium fuel-cycle technologies for self-sufficiency in future fusion operations.

It's also crucial for STEP Fusion, the UK’s flagship initiative to demonstrate fusion energy as soon as the 2040s – which the government claims will make the UK the first country in the world with a clear path to commercial fusion that could ensure lasting energy security.

"With Sunrise, the UK will have a powerful new capability to rapidly and accurately simulate plasma behaviour and fusion conditions, helping researchers advance the development of stable, efficient and economically viable fusion energy," said Thomas Zacharia, SVP, strategy and development, public sector, AMD.

"Sunrise brings together AMD EPYC processors and AMD Instinct GPU acceleration, purpose-built on the Dell PowerEdge platform, to deliver breakthrough AI and high-performance computing for the UK fusion community, and supporting the Department for Energy Security & Net Zero as it moves fusion from research toward practical impact."

According to the government, Sunrise is set to be the world’s most powerful AI supercomputer dedicated to fusion energy, and is scheduled to come into operation this June.

"We can be proud that Britain will lead the way on research, innovation and skills for a future of limitless fusion energy," said minister for science, innovation, research and nuclear Lord Vallance.

"By backing our fusion industry, we are not only securing our future energy independence, but from innovation and research to engineers, we are also providing the skilled clean energy jobs of the future for British people."

FOLLOW US ON SOCIAL MEDIA

“Agent Computers” mark the next evolutionary step in the PC market, according to AMD, with the chipmaker predicting a wave of agentic AI–infused devices in the near future.

In a recent blog post, AMD said "AI on the PC has crossed an important line", noting that there's a growing shift toward running agents locally on hardware, particularly with the rise of OpenClaw.

Developed by Peter Steinberger, OpenClaw is an open source agent designed specifically to run on a laptop or PC and essentially act as a digital assistant.

Users can order agents to carry out tasks on their behalf, such as sending emails, managing appointments, or clearing out inboxes.

This represents a step change in how users interact with on-device AI, according to AMD, marking a shift away from ambient capabilities and actively commanding assistants toward more autonomous AI-powered activities.

“The PC was built for direct human operation. You open apps, issue commands, and work step-by-step,” the chipmaker noted. “Agents work differently. They run continuously, handle multiple tasks in parallel, and move across tools autonomously.”

“An Agent Computer is a new category of device built to run your AI agents full-time,” the blog post added. “It can sit in your home or office, always on, always available, always working. You do not operate it like a PC. You delegate to it.”

Notably, AMD suggested that interoperability with third-party applications will be a key differentiator for this new wave of agent computers, allowing users to command agents via platforms such as WhatsApp or Slack.

“You send a message on WhatsApp. Your agent gets moving,” the company wrote. “You drop a task into Slack. Your agent takes it from there.”

A vested interest in Agent Computers

It’s worth noting that AMD has a vested interest in championing the prospect of Agent Computers. Indeed, the company touted its own range of AI Max processors as an ideal option for devices if this trend takes off.

The company said systems powered by AMD Ryzen AI Max+ 395 range of processors are “exceptionally well suited to become Agent Computers”.

“They have the horsepower to run sophisticated local models, the efficiency to stay persistently available, and the architecture to support the parallel, multi-agent workloads that define what comes next,” the company said.

The next step in on-device AI?

AMD’s position here appears to rest on drawing comparisons with broader generative AI advances over the last three and a half years.

The early days of the generative AI “boom” saw big tech providers roll out an array of AI “assistants” and “copilots”, yet the focus has since shifted to agentic AI. As ITPro reported around the launch of Salesforce’s Agentforce service, CEO Marc Benioff described agents as the next evolutionary step in the trajectory of enterprise AI.

AI PCs have been an area of intense focus for hardware manufacturers over the last two years, with a plethora of devices now infused with built-in AI features.

In September last year, Gartner predicted that AI PCs will “become the norm” by 2029 as enterprises and consumers alike flock to these powerful new machines. Going by Benioff’s logic of AI evolution, AMD appears to be betting big that this process is now unfolding in the device market.

AMD noted these devices have applications across both the enterprise and consumer markets, helping to reduce manual toil or even create more convenient browsing experiences.

“For professionals, an Agent Computer means more output and more leverage,” the company noted. “For creators, more time for original work and less time managing logistics. For developers, a local AI environment purpose-built for building, testing, and running agents.”

“An Agent Computer does not replace your abilities. It amplifies them.”

FOLLOW US ON SOCIAL MEDIA

AMD CEO Lisa Su has revealed AI is directly influencing recruitment practices at the chip maker but, unlike some tech firms, it’s led to increased headcount.

Speaking to CNBC at the CES 2026 conference in Las Vegas, Su said prospective hires with AI expertise and experience are now a priority as the company targets further gains in this sector.

“We’re growing very significantly as a company, so we actually are hiring lots of people,” Su said. The profile of these new recruits has changed, however, with the company hiring “people who are AI forward”.

Su added that the chip maker is now using AI to support development and testing of chips, which is having a tangible positive impact on workforce productivity and efficiency.

“I would say that AI is augmenting our capabilities,” she told CNBC. “It’s not replacing people, it’s actually just augmenting our productivity in terms of the number of products we can bring up at any given time.”

The AMD chief’s comments come in the wake of a year fraught with concerns about the impact of AI on the global labor market. Industry stakeholders have issued repeated warnings that the technology could adversely affect entry-level workers and long-term talent pipelines.

IT, customer support, and software development have all been touted as areas in which AI has the potential to take human jobs, with comments from major industry figures only intensifying concerns.

In early 2025, Salesforce CEO Marc Benioff suggested the company may reduce hiring in software engineering as a result of gains made with its own in-house agentic AI service, Agentforce.

Elsewhere, Anthropic CEO Dario Amodei warned AI could wipe out nearly half of all entry-level roles within five years.

AMD isn’t alone in AI hiring changes

AI is influencing hiring at enterprises globally, with the trend expected to continue in the coming years. This disruption might be more positive than some critics believe, however.

Analysis from Gartner in November 2025 predicted around 32 million jobs a year will be “reconfigured, redesigned, or fused” from 2028 onward.

The consultancy told ITPro at the time that a “jobs apocalypse” is unlikely with a long-term net positive gain on job creation anticipated.

Notably, a December survey of UK business leaders from project management company Monday seemed to back this prediction up, with more than three-quarters (78%) of respondents saying they don’t expect a drop in hiring across 2026.

Nearly one-third (32%) of respondents even said they expect to hire more staff this year as a result of their AI strategies.

FOLLOW US ON SOCIAL MEDIA

France is set to deploy its first exascale supercomputer following a deal between AMD and Eviden, part of Atos Group.

Named Alice Recoque after the pioneering French computer scientist, the new exascale system will serve as an ‘AI Factory’ operated by the French Alternative Energies and Atomic Energy Commission (CAE).

Development of the supercomputer is set to cost €554 million ($638.85 million), with investment secured via the European Union’s EuroHPC Ju initiative and additional support from the Digital Europe Programme (DEP and the Jules Verne Consortium).

In an announcement confirming the project, Dan McNamara, senior vice president and general manager for compute & enterprise AI at AMD, said the new system will help drive the “next generation of innovation” in AI and high-performance computing (HPC).

“The Alice Recoque supercomputer represents a major step forward for European sovereign AI, uniting national ambition, regional collaboration, and AMD’s high-performance and AI compute technologies,” he said.

“Through our continued collaboration with EuroHPC JU, the Jules Verne Consortium, and Eviden, we are proud to support Europe’s scientific and industrial leadership with a platform purpose-built for scale, efficiency and discovery."

Under the hood of the Alice Recoque supercomputer

The Alice Recoque supercomputer will be equipped with next generation AMD “Venice” EPYC CPUs, unveiled in April this year, alongside AMD Instinct MI430X GPUs.

Each GPU boasts 432 GB of HBM4 memory alongside 19.6 Tbit/sec of bandwidth, AMD revealed.

Elsewhere, a new MI400 Series accelerator will power the system, which AMD said has been specifically engineered for “sovereign AI and scientific computing”.

Underpinning AMD's hardware is Eviden’s BullSequana XH3500 high-performance supercomputing platform, including its Argos software to optimize energy efficiency.

All told, the duo said Alice Recoque will be capable of reaching exascale computing speeds with “fewer resources and reduced energy consumption rates”, establishing it as one of the most efficient systems in Europe.

Indeed, the system uses 25% fewer racks and components compared to other exascale systems, also boasting 50% better energy efficiency rates per GPU.

“Composed of 94 racks, Alice Recoque is expected to be one of the top supercomputers in Europe for double-precision HPC workloads,” AMD said in a statement. “It will also offer exceptional memory performance enabling deeper insights, faster simulations, and more scientific breakthroughs.”

“Eviden’s architecture will enable Alice Recoque to deliver maximum performance at minimum cost and power, to meet Europe’s demanding green computing goals.”

What will the supercomputer be used for?

Once operational, the new exascale system will be used to tackle some of Europe’s “most pressing societal, scientific, and industrial challenges”, according to AMD.

This will include projects focusing on climate modelling, digital twin simulations, advanced materials research, and development of next-generation AI models.

Emmanuel Le Roux, group SVP and global head of advanced computing and AI at Eviden, said the new system marks “another critical step” in bolstering European supercomputing capabilities.

“As a catalyst for scientific and industrial breakthroughs, from climate modeling and healthcare to advanced materials and AI innovation, it will empower researchers and industries across Europe,” Le Roux commented.

“Born from a shared European vision, this AI Factory reflects on what we can achieve collectively toward a common goal. Eviden is fully dedicated to its success, bringing deep expertise, a collaborative spirit, and a long-term dedication to responsible technological leadership.”

The announcement marks the latest in a string of supercomputing announcements from AMD in recent years. Hardware developed by the company currently powers 40% of the top 10 fastest supercomputers in the world.

This includes El Capitan, housed at the Lawrence Livermore National Laboratory, as well as Frontier, the world’s first exascale system housed at Oak Ridge.

MORE FROM ITPRO

• Inside Isambard-AI: The UK’s most powerful supercomputer
• Nvidia just announced new supercomputers and an open AI model family at SC 2025
• HPE's new Cray system is a pocket powerhouse

In a world where many folk will baulk at even installing their operating system on a PC, asking them to also undertake some physical assembly may seem a foolhardy idea. The small amount of basic work you need to undertake with a screwdriver should, however, not deter you from buying the new Framework Desktop PC.

Framework is not a new player in the PC market, though its previous products have been laptops rather than desktops. While the new Framework Desktop is quite a different beast, you can still use the same slide-in modular expansion cards that feature on its current laptop models.

You can order your Framework PC with either Windows 10 Home or Pro, or as a barebones system without an operating system. Either way, you will need to install your OS, as the Windows option simply means you get an activation code from Framework.

We loaded our review machine with Ubuntu 20.04 and Windows 11 Pro in a dual-boot setup, even though Framework recommends Fedora and Bazzite as officially supported Linux distros.

It also recommends Arch, NixOS, and Bluefin as community-supported distros, but the fact that Ubuntu loaded and worked with no issues suggests most versions of Linux will run without any problem.

Framework Desktop: Design

Once you've unboxed the various components that make up the Framework PC package, you are left with a 4.5L metal PC tower, which already includes the motherboard, heat sync, power supply, and all the other essential gubbins, as well as the bits you need to install yourself and a Torx screwdriver.

These include a fan (from a choice of three), one or two M.2 2280 SSDs, assuming you've specified them, clip-on plastic side panels for the tower, and various decorative plastic tiles to clip into the magnetic front panel.

You also receive any modular expansion ports you've ordered, which slide into the two docking bays at the bottom of the tower. They are the same as the ones you'll find on Framework's laptops. In our case, this was a selection of USB-C and USB-A expansion cards.

The assembly part of the process involves installing the SSD and the fan, then the side panel, the two expansion ports, and the decorative tiles. Framework reckons this all takes 10 minutes, but we timed ourselves and got it done in under 8. It took longer to unbox all the parts and clear the mess away.

If you want a get a better taste of what is required in the assembly process, Framework provides a selection of foolproof guides to putting the tower together and installing your software.

While assembling the unit, it became clear that Framework hasn't skimped on the quality. Everything went together smoothly and securely, all the clips and screws felt well-made, and the result felt reassuringly stiff and sturdy.

Adding a second SSD takes even less time: all you need to do is slide out the plastic panel that covers the main board and screw it into place. The Framework can take SSDs up to 8TB in capacity, making it a great choice if you want to store very large quantities of data.

Framework offers a wide choice of expansion cards for the two slots at the bottom of the unit. You can choose from 10Gbps USB-C or USB-A I/O ports, MicroSD or SD card readers, 2.5Gbit Ethernet, and 3.5mm audio jacks and storage cards with either 256GB or 1TB capacity. The cards vary in price from £9 for the USB-C/A cards up to £39 for the Ethernet card and £45/£125 for the storage modules.

The two expansion slots on the front of the tower are in addition to a wide selection of fixed ports at the back, namely two USB-C 4 and two USB-A 3.2 Gen 1 ports, two DisplayPort 2.1 and one HDMI 2.1 video outputs, 5Gbit Ethernet, and 3.5mm audio jacks.

All together, that makes the Framework Desktop one of the best-connected small-form PC we've ever seen here at ITPro. Wireless communications are handled by a Realtek RZ717 card supporting the latest Wi-Fi 7 and Bluetooth 5.4 standards.

The slots for the expansion cards are at the very bottom of the case, and this is the only small problem with the design because it puts the ports only a few millimetres above the surface the Framework is resting on, which isn't ideal for ease of access, especially for bulky USB thumb drives.

If you have to pick the box up to use the expansion ports, you'll thank the optional screw-on carry handle that fixes to the top of the tower. It's not strictly necessary, given that the Framework PC only weighs 3.1Kg, but it does make grabbing it easier.

The magnetic front panel can be decorated with 21 plastic tiles. These are available in packs of 7 coloured tiles for £15, so you have to spend £45 just to fill all the spaces. For £5 each, you can buy individual tiles with various decorations such as the AMD, Tux, Framework, and Linux Mint logos. Of course, nothing is stopping you from making your own tiles if you have a 3D printer to hand.

Given that the unit looks half-finished without a full set of tiles to fill the front panel, and the £45 cost for something purely cosmetic, perhaps Framework could have considered also including a simple plastic cover for the front?

Framework Desktop: Specs and Performance

The Framework Desktop is the first PC we've encountered running on AMD's new Halo Strix Max CPUs. You can have your Framework Desktop with either an 8-core AMD Ryzen AI Max 385 and 32GB of RAM or the 16-core Max+ 395 CPU with 64GB or 128GB (the RAM is integrated, so there's no option to add more post-purchase), along with a 32-core Radeon 8050S or 40-core 8060S GPU. Our test unit came with the basic 385 CPU, but even that proved to be surprisingly powerful.

In ITPro's in-house 4K multi-media benchmark, the Framework scored a blistering 510 points, while the PCMark10 test score came in at 8,856. If you want a small-form PC with meaningfully higher levels of performance, you'll need to spend twice as much or more on something like a M4 Max Mac Studio.

To put the Framework's performance into context, the new Core Ultra 288V-powered MSI NUC AI+ 2MG scored 301 points in the 4K test, while the Core Ultra 9 285H-powered Geekom IT15 managed 361.

In the PCMark10 benchmark, the NUC AI+ 2MG scored 6,857 and the Geekom box 7,450 points. The result is clear: when it comes to demanding workloads, the new Framework Desktop is in a league of its own at this price point.

Graphics performance was, if anything, even more impressive, with the SPECviewperf 3dsmax 3D modelling test running at 118fps, which is close to five times as fast as the 25fps you'll get from one of Intel's latest Arc integrated GPUs.

The Framework Desktop is not being pitched as a gaming platform, but it seemed churlish not to test its gaming chops. The Cyberpunk 2077 benchtest ran at 73fps at FullHD in the Ray Tracing: Medium preset. That's very impressive.

In more general terms, the Ryzen AI Max 385 chipset delivers GPU performance that is roughly comparable to an Nvidia 4060/4070 GPU with an 80-100 watt TGP. That's an order of magnitude more powerful than any iGPU we've come across, including the likes of AMD's 890M and Intel's Arc 140T.

Perhaps even more impressive than the outright performance is the lack of fuss with which the Framework Desktop can handle even the most demanding tasks. During our standard stress test, when we ran the CPU and GPU flat out, the system was still running at 100% utilization after three hours, and the fan noise was very subdued.

The 1TB Western Digital Black SSD in our review unit proved to be a strong performer, recording average sequential read and write speeds of 5,020 MB/s and 3,240MB/s, respectively.

Framework Desktop: Is it worth it?

The system we were sent to test with a single Western Digital Black 1TB SSD, the basic fan, and two expansion cards, but no Windows licence will set you back £1,226. A Windows activation licence will cost a further £120 for Home or £220 for Pro if bought from Framework, but many of you reading this may already have access to a corporate or educational Windows licence.

With or without a Windows licence, the Framework Desktop certainly offers a serious amount of performance for the asking price. Indeed, it's hard to argue with the basic offer of a small form PC with power enough to run even the most demanding graphics jobs quickly and demanding 3D games for under £1,500.

The modular expansion ports are perhaps not as important in the Desktop as they are in Framework's laptops because even without them, the Desktop still has a good selection of ports, but they are still a useful feature.

We'll leave it to the reader to decide if the decorative tiles are essential parts of the Framework Desktop concept or more of an aesthetic (and expensive) indulgence, but given the price, performance, and quality of the overall package, we like that the Framework Desktop doesn't look like just another boring black boxy small-form PC.

Framework Desktop Specifications

AMD's Advancing AI conference is one of the shortest flagship events I've been to, certainly this year and possibly ever, but still managed to be fairly action-packed.

The one day event was centered around CEO Lisa Su's two-hour keynote presentation, in which she was joined on stage by numerous special guests – including a surprise appearance from OpenAI CEO Sam Altman.

I'll return to these guests in a moment, but first let's dig down into the news announcements. The biggest news was Helios, a double-wide, rack scale offering from AMD that will be available next year, which Su described as a "game changer" for the company.

To say there was a lot of pointing to graphs where AMD's devices either match or exceed Nvidia's comparable offerings would be an extraordinary understatement. In Su's keynote we got to see a slide claiming that the Instinct MI355X chip – also announced at the event – is 1.2x faster at DeepSeek R1 throughput and 1.3x faster at Llama 3.1 405B throughput than Nvidia's B200, and offers the same throughput as the Nvidia GB200 on Llama 3.1 405B. This was followed by a claim that the MI355X offers 'up to' 40% more tokens per dollar than the B200.

This was followed by similar slides showing parity or slight gains on LLM training and fine tuning, while in the press conference we were treated to further slides showing greater memory capacity, memory bandwidth, and peak performance than Nvidia GB200 or B200.

As we're not able to independently verify these claims, I'm not going to repeat them all verbatim, but needless to say there was a strong "better than Nvidia, actually" theme.

In many ways this is to be expected. Like it or not, AMD and Intel are both playing catch-up with Nvidia, a company led by perhaps one of the most charismatic CEOs out there.

As Forrest Norrod, EVP and GM of AMD's data center solutions business unit, told a press round table: "Nvidia has done a great job of convincing the industry that we're living in the future, and so that whatever... they announce is actually here today, so we have to get a little bit more aggressive."

Norrod also admitted that "Nvidia is the de facto standard right now," but added that Helios will be a "good solution" for hyperscalers, tier two cloud providers, and neo clouds alike. He was also at pains to point out that Helios is "not the only thing [AMD is] doing", however, and while he didn't expand on this previous slides had shown plans for future rack-scale architecture going into 2027, as well as an annual chip release plan going up to the Instinct MI500X range. Watch this space, I suppose.

Embracing the open ecosystem

The final key talking point for Su and other members of the AMD leadership was the importance of openness. Su said during her keynote that AMD is "investing heavily in an open, developer-first ecosystem," adding that the company is "really supporting every major framework, every library, and every model to bring the industry together in open standards so that everyone can contribute to AI innovation".

Su drew a parallel between the self-declared open approach that AMD is taking and Linux surpassing Unix as the data center operating system of choice "when global collaboration was unlocked". She also pointed to the open source Android operating system as one of the drivers in increasing the number of people who own smart phones. Indeed, the latest figures from Statista at the time of writing show Android had a 72.7% market share, while iOS is just 26.9%, despite iPhones being the most popular device.

As mentioned above, many of the graphs in Su's presentation leaned on the throughput performance of Llama – Meta's ostensibly open source AI platform – and Vamsi Boppana, SVP of AI at AMD, pointed out that 1.8 million Huggingface models now run on the company's ROCm platform.

What's next for AMD

It's hard to say when the next AMD Advancing AI event will take place – typically companies this size run their conferences annually, but only nine months had passed between the 2024 event and this most recent one. Nevertheless, the company has given a clear indication of what its plans are for the next few years at least.

Su committed to an annual release cycle for its Instinct GPU range, which is unlikely to be broken. We can expect Helios to ship in 2026 and a second rack-scale product to come in 2027. What the full specs of all those upcoming products are – and how much interest there is in Helios in particular – remains to be seen. One thing is certain, though – Nvidia may be the "de facto standard" but AMD's spoiling to take its crown.

AMD has taken the wraps off a new, fully integrated, rack-scale AI infrastructure solution aimed at both frontier model training and large-scale AI inference.

The product, dubbed Helios, was revealed by AMD CEO Lisa Su at the company’s latest Advancing AI conference in San Jose, California.

“Helios is truly a game changer,” said Su during her keynote at the event. “For the first time, we've architected every part of the rack as a unified system that's combining our CPUs, our GPUs, our Pensando NICs and our ROCm software all together in one platform, and it's really purpose built for the most demanding AI workloads.”

Speaking to press ahead of the launch, Andrew Dieckmann, CVP and GM for AMD’s data center GPU business, said: “[Helios] will set the new standard for rack scale AI (and) we expect Helios to deliver significant advantages over the best 2026 rack scale solutions from our competition.”

“We have competitive flops, we have 50% more HPM memory capacity, we have 50% more HPM memory bandwidth, and up to 50% more scale our bandwidth,” Dieckmann continued.

“These advantages translate directly into faster model training and better inference performance and hugely advantaged economics for our customers.”

Under the hood of the Helios AI rack

Helios, when it ships in 2026, will feature AMD’s latest and greatest chipsets. The Instinct MI400 GPU series, also available in 2026, is – according to Dieckmann – the most advanced accelerator the company has ever built.

“It’s designed to run hundreds of millions and trillion parameter models and train them and provide inferencing on them.

It features 40 petaflops of FP4 performance, 20 petaflops of FP8, 432 gigabytes of HBM per GPU and supports almost 20 TB/sec of HBM memory bandwidth. [Each chip] also supports up to 300 GB/sec of scale-up bandwidth, enabling high bandwidth interconnect across racks and clusters,” said Dieckman.

Aside from the MI400, the system also features upcoming AMD Zen 6-based Epyc Venice CPUs and AMD Pensando Vulcano network interface cards (NICs), all of which are coming in 2026 as well.

Delegates at Advancing AI also got a brief glimpse of AMD’s future AI plans, with the company moving to an annual cadence of AI infrastructure releases.

In 2027, the company plans to release a “next-gen AI rack” that will also feature AMD Pendando Vulcano NICs, as well as Epyc Verano – the successor to Venice – and MI 500 Series Instinct GPUs.

MORE FROM ITPRO

• AMD eyes networking efficiency gains in bid to streamline AI data center operations
• European AI alliance looks to take on Silicon Valley and develop home-grown LLMs
• AMD just made a big statement in the AI PC race with its Ryzen AI Pro 300 series processors

Welcome to ITPro's coverage of AMD Advancing AI 2025, coming to you live from the San Jose McEnery Convention Center. Hundreds of delegates are expected to fill the auditorium today to hear AMD CEO Lisa Su reveal the company's latest strategic plans for AI.

Su is expected to take to the stage at 9.30am PST and with her keynote slated to last for two hours, we can undoubtedly expect guests from both AMD and its partners to grace the stage.

With a little under two hours to go before Lisa Su's keynote starts, it's worth reflecting on what happened at the company's last Advancing AI conference, which only took place eight months ago in October 2024.

At that event, Su launched the enterprise-focused Ryzen AI Pro 300 Series CPU range, revealed the Pensando Selina data processing unit, and announced the arrival of Instinct MI325x accelerators. It's unlikely that the company has invited customers, partners, and press to hear updates on these products, but it could indicate the direction of travel.

At Computex in June 2024 Su announced the company was moving to an annual release cadence for its GPUs and with the launch of the Instinct MI325x accelerator series at the last Advancing AI we could see an update in this area.

The auditorium is starting to fill up at the San Jose McEnery Convention Center as we wait for Lisa Su to take to the stage for her keynote address this morning.

And we are underway with an opening video – apparently the key for the AI future is trust.

And here is the woman of the hour, Dr Lisa Su, CEO of AMD.

"I'm always incredibly proud to say that billions of people use AMD technology every day, whether you're talking about services like Microsoft Office 365 or Facebook or zoom or Netflix or Uber or Salesforce or SAP and many more you're running on AMD infrastructure," says Su. "In AI, the biggest cloud and AI companies are using instinct to power their latest models and new production workloads. And there's a ton of new innovation that's going on with the new AI startups."

As with many companies now – including partner Dell and rival Nvidia – AMD is talking up AI agents and its (positive) impact on hardware.

Su says that agentic AI represents "a new class of user".

"what we're actually seeing is we're adding the equivalent of billions of new virtual users to the global compute infrastructure. All of these agents are here to help us, and that requires lots of GPUs and lots of CPUs working together in an open ecosystem," she says.

AI isn't just a cloud or data center question now, it's also an endpoint question, and particularly the PC. "We expect to see AI deployed in every single device," she says.

Su hits on another theme that's becoming popular among many enterprise hardware providers: open source and openness.

"AMD is the only company committed to openness across hardware, software and solutions," Su claims. "The history of our industry shows us that time and time again, innovation truly takes off when things are open."

"Linux surpassed Unix as a data center operating system of choice when global collaboration was unlocked," she continues. "Android's open platform helped scale mobile computing to billions of users, and in each case, openness delivered more competition, faster innovation and eventually a better outcome for users. And that's why for us at AMD, and frankly, for us as an industry, openness shouldn't be just a buzzword. It's actually critical to how we accelerate the scale adoption and impact of AI over the coming years."

Su is now joined on stage by her first guest, Xia Sun of xAI, to talk about how the company behind Grok is working with AMD.

Sun focuses on how his small team is being bolstered by AMD hardware like MI300x and – of course – openness

Sun leaves the stage and now we have our first product announcement, the AMD Instinct MI350 Series.

"With the Mi 350 series, we're delivering the largest generational performance lead in the history of instinct, and we're already deep in development of Mi 400 for 2026," says Su.

"Today, I'm super excited to launch the Mi 350 series, our most advanced AI platform ever that delivers leadership performance across the most demanding models this series," she says, adding that while AMD will talk about the MI355 and the MI350 "they're actually the same silicon, but MI355 supports higher thermals and power envelopes so that we can even deliver more real world performance."

Su claims that the MI350 Series "delivers a massive 4x generation leap in AI compute to accelerate both training and inference". Some of the key claimed specs are:

• 288GB memory
• Running up to 20 billion paramaters on a single GPU
• Double the FP throughput compared to competitors
• 1.6x more memory than competitors

(These are AMD claims that haven't been independently verified by ITPro)

And now another guest: Yu Jin Song, Meta VP of engineering – Meta, of course, being the force behind Llama LLM.

"We're seeing critical advancements that started with [inferencing[ and now extending to all of your AI offering," says Yu Jin Song. "I think AMD and Meta have always been strongly aligned on ... neural net solutions."

"We try to shoot problems together, and we're able to deploy optimized systems at school. So our team really see AMD as a strategic and responsive partner and someone that we really like. Well, we love working with your engineering team. We know that we love the feedback, and you also hold the high standard. You are one of our earliest partners in AI," he adds.

Yu Jin Song explains that MI350 will become a key part of the company's AI infrastructure (it's already using MI300x accelerators).

"We're also quite excited about the capabilities of MI350x," he says. "It brings significantly more compute power and acceleration memory and support for FP4, FP6 all while maintaining the safety pull factor at mi 300 so we can deploy quickly."

These are some of the Instinct MI350 Series partners, which will have products coming to market with the chips in Q3 2025, including Oracle, Dell Technologies, HPE, Cisco, and Asus.

We're going from hardware to software, now, with Su saying this is "really [what] unlocks the full potential of AI". She leaves the stage and Vamsi Boppana, SVP of AI at AMD, takes the metaphorical mic to talk about ROCm.

"We have been relentlessly focused on what matters most, making it easy for developers to build better out of the box capabilities, easy setup, more collateral, stepping up community engagements," he says.

"We've been running hackathons, contests, meetups and mail, and our customers are deploying AI capabilities at unprecedented pace, and that's why we've significantly accelerated hour release cadence," he continues. "New features and optimizations are now shipping every two weeks. Leading models like llama and DeepSeek work on day zero."

He also points out AMD's partnerships with Hugging Face, Pytorch, OpenAI Triton, and Jax.

Now for his announcement: ROCm 7. This will, he says, "bringing exciting new capabilities to address ... emerging trends, and brings support for our MI350 series of GPUs".

"[It] introduces the latest algorithms, advanced features like distributed inference, support for large scale training and new capabilities that make it easy for enterprises to deploy AI effortlessly within ROCm 7," he says, adding that inference has been "the largest area of focus [and] we've innovated and invested at every layer of inference".

Boppana now invites his own guest to the stage, Eric Boyd CVP of AI platforms at Microsoft.

The two are discussing Microsoft and AMD's longstanding partnership, with Boyd saying Microsoft has been using "several generations of Instinct".

"It's been a key part of our inferencing platform, and we've integrated ROCm to our inferencing stack, making it really easy for us to take and deploy the models on the platform," he says.

Boppana is now talking about what he describes as a new opportunity for driving down inference costs: distributed inference.

"In any LLM serving application, there are two phases. There's the pre fill phase and there's decode phase," explains Boppana. "These two phases of the model are typically handled on the same GPU, but now if you [use] the same GPU, it often becomes a bottleneck for large models, or when demand spikes happen and you can get limited in performance or flexibility."

Disaggregating the pre fill and decode phase can "significantly improve throughput, reduce cost, and boost responsiveness", he says.

Here is that solution: ROCm Enterprise AI. "ROCm Enterprise AI makes it easy to deploy AI solutions with new cluster management software, it ensures reliable, scalable and efficient operation of AI cluster," says Boppana.

"Our mlops platform allows fine tuning and distillation of models with your own data and a growing catalog of models will come for specific industries," he adds.

Another announcement from Boppana: AMD Developer Cloud and Developer credits. All developers who are here will have a credit voucher in their email inbox right now, he says, which elicits cheers from the assembled crowd.

Boppana hands over to Anush Elangovan, CVP of software development, and Sharon Zhou, CVP of AI. Elangovan channels Steve Balmer: "Developers! Developers! Developers!" to yet more cheers – although he stopped after just three.

"We are serious about bringing rocket everywhere and to everyone, from client to the cloud," he continues. "In AI, speed is your mode. Access to compute is paramount. We've been delivering on speed, so now let's get you access to compute."

"With the developer cloud, anyone with a GitHub ID or an email address can get access to an instant GPU with just a few clicks," he says, directing developers who want to give it a go to dev.amd.com.

"We've also included a lot of easy to use frameworks like vlln, SG, Lang, pytorch, etc," he says. "You just select one of those frameworks, add your SSH key and then create and you're set."

Zhou briefly encourages developers in the crowd to "come say hi" at hackathons, university labs and elsewhere, before handing back to Boppana, who reveals the next of today's announcements: The ability to access ROCm 7 on Windows or Linux, on device or on client.

Boppana hands over to Forrest Norrod, EVP and GM of data center solutions, who is underlining the importance of agentic AI for AMD, as well as the performance improvements of fifth-generation Epyc CPUs.

He also talks up what AMD says is the continued importance of CPUs.

"Some will naively tell you that CPUs are less important in the age of AI, but that's not correct," he says. :With agentic AI, we see an explosion of autonomous agents accessing data and enterprise applications. This increases the needs for efficient, high performance x86 compute across the data center, then within the AI server itself, the CPU serves the demands of pre processing and workload orchestration to keep the GPUs working efficiently."

"Let me show you a few examples of how the right CPU can manage GPUs work better," he says. "Choosing poorly can create bottlenecks and stranded valuable resources. As you can see across a range of models and use cases, our fifth generation epic CPUs can boost the inference performance of the entire system from six to 17% that makes a huge impact on the overall TCO and performance of The AI deployment, and it's a critical element in designing the best possible AI system. So get much more out of your GPUs with the right CPU."

Lisa Su is back on stage to give the audience "a first look at the next big step for [AMD's] AI roadmap, the MI400 series". She quickly pivots from there to introduce Helios AI Rack.

"Think of Helios as really a rack that functions like a single, massive compute engine," Su says. " It connects up to 72 GPUs with 260 terabytes per second of scale up bandwidth. It enables 2.9 exaflops of FP four performance, and that is a great number."

Naturally, Su also takes a moment to make comparisons with AMD's competition, saying Helios supports "50% more HBM ... memory bandwidth, and scale out bandwidth".

And now, a surprise guest, "someone who's really an icon in AI", says Su: Sam Altman, CEO of OpenAI.

Altman talks about his company is seeing in the market.

"I think the models have gotten good enough that people have been able to build really great products, text, images, voice, all kinds of reasoning capabilities," he says.

"We've seen extremely quick adoption in the enterprise now, coding has been one area people talk a lot about, but I think what we're hearing again and again in all these different ways, is that these tools have gone from things that were, you know, fun and curious to truly useful," he says.

Asked by Su how compute demands are changing, Altman says: "One of the biggest differences has been, we've moved to these reasoning models. So we have these very long roll outs where a model that often think about come back with a better answer.

"This is really the pressure on model efficiency and long context for all this, we need tons of computers, tons of memory, tons of CPUs as well. And our infrastructure ramp over the last year and what we're looking over the next year has just been a crazy, crazy thing to watch."

Turning to the hardware used to support OpenAI's platforms like ChatGPT, Altman says the company is already running some work on the MI300x, "but we are extremely excited for the MI450, the memory architecture is great for inference. It can be an incredible option for training as well".

Su asks Altman what he sees in the future for AI. "How will the workloads evolve? What happens with – quote, unquote – AGI (artificial general intelligence)," she says.

"I think we're going to maintain the same rate of progress rated models for the second half of the decade as we did for the first," Altman responds. "I wasn't so sure about that a couple of years ago for a new research thing to figure out, but now it looks like we'll be able to deliver on that."

"So if you think forward to 2030 ... these systems will be capable of remarkable new stuff, about scientific discovery, running extremely complex functions throughout society, and things that we just couldn't even imagine," he continues. "It's really going to take these are huge systems now, very complex engineering projects, very complex research, and to keep on this [path of] scaling, we've got to work together across research, engineering, hardware, how we're going to deliver these systems and products.

"This has gotten quite complex, but if we can do on that, if we if we can deliver on that drives collaboration across the whole industry, we will keep this curve going," he concludes.

Altman leaves the stage and Su delivers some final thoughts to the audience.

"When I think about this past year, it's really redefined what progress in AI looks like. It's really moved at a pace unlike anything that we have seen in modern computing ... anything that we've seen in our careers, and, frankly, anything that we've seen in our lifetime," Su says.

"The future of AI is not going to be built by any one company or in a closed ecosystem. It's going to be shaped by open collaboration across the industry. It's going to be shaped because everyone is bringing their best ideas, and it's going to be shaped because we're innovating together. So on behalf of all of us at AMD, we look forward to changing the world with you together," Su concludes, and with that the keynote is over.

Computer hardware and processor manufacturer AMD has announced the appointment of James Blackman as head of VAR and SI commercial sales for the EMEA region.

A seasoned industry veteran, Blackman brings more than 30 years’ experience to the role and has spent more than 13 years at AMD, most recently leading the firm’s HP sales activity across EMEA.

In his new position, the sales veteran is now tasked with building a robust channel community in the region, while acting as a clear point of contact for channel partners, the chipmaker said.

“I'm delighted to lead the AMD VAR & SI Commercial Sales team in EMEA,” Blackman said in an announcement.

“My focus will be on strengthening our partner ecosystem to deliver exceptional value to our customers. I recognize the immense market opportunity for AMD solutions, and I'm eager to collaborate with ITDMs to achieve mutual success.”

Blackman will leverage his extensive experience and expertise to support partners as they leverage AMD technologies across various value added resellers (VARs) and systems integrators (SIs) - including its high performance compute capabilities, sustainability innovations, and AI offerings.

The firm said the appointment underscores its commitment to strengthening its channel partnerships and expanding its reach within the EMEA region, with Blackman to “play a key role” in helping partners deliver solutions to their customers.

“In my previous role, I had the privilege of partnering exclusively with HP. Now this new position allows me to embrace the full AMD portfolio and collaborate with all OEMs across the ecosystem,” Blackman said in a LinkedIn post.

“It’s a fantastic opportunity to drive growth, strengthen partnerships, and work closely with value-added resellers and systems integrators across EMEA.”

The appointment marks the latest move by AMD to strengthen its sales and channel leadership across EMEA, with the company naming HP veteran Stephanie Dismore as its new senior vice president of EMEA regional sales back in August 2024.

Elsewhere, the chipmaker also recently announced its latest earnings report, which revealed record quarterly revenue of $7.7 billion for Q4, alongside a company best of $25.8 billion for the entirety of its FY2024.

MORE FROM ITPRO

• GPU cloud startup Vultr secures AMD backing in $333 million investment round
• Researchers claim an AMD security flaw could let hackers access encrypted data
• AMD to cut around 1,000 staff to focus on "growth opportunities"

Cloud infrastructure company Vultr has raised $333 million in funding as part of an investment round that values the firm at $3.5 billion.

Founded in 2014 by David Aninowsky and self-funded until now, Vultr operates a global, full-stack AI infrastructure and cloud computing platform with 32 cloud data center regions around the world.

Led by LuminArx Capital Management and AMD Ventures, the investment will enable the firm to accelerate its global expansion and capitalize on the surge in enterprise AI compute demands.

"Building upon the foundation created by David Aninowsky, Vultr has become the leading cloud infrastructure alternative to traditional hyperscalers,” said JJ Kardwell, CEO of Vultr.

"As we continue our global expansion, LuminArx’s collaborative, innovative approach and AMD’s strategic impact enable Vultr to further accelerate growth in AI infrastructure and cloud computing, creating the category-defining independent cloud infrastructure company."

Vultr is one of a number of startups aiming to capitalize on the generative AI boom by offering alternative GPU and infrastructure options to industry giants such as Nvidia.

Earlier this year, another GPU infrastructure provider CoreWeave raised $1.1 billion in funding and secured multi-billion dollar deals with Microsoft and Cisco.

Analysis from 451 Research - and commissioned by Vultr - showed that enterprises will push beyond AI training to the edge in 2025, presenting an opportunity for startups in this domain to capitalize on the trend.

This latest investment will enable the firm to expand its global cloud GPU fleet to support this widespread industry shift.

"Vultr’s deep experience delivering secure, compliant, and scalable cloud infrastructure and their deployment of AMD Instinct accelerators positions them as an innovative cloud solutions provider," said Mathew Hein, senior vice president and chief strategy officer of corporate development at AMD.

"We share Vultr’s mission to empower enterprises and AI innovators with unparalleled access to high-performance compute for AI model development and deployment and are proud to support their growth."

The investment round marks the latest seal of approval by major industry players.

Earlier this month, Vultr signed a four-way deal with Juniper Networks, Broadcom, and AMD to launch a new network architecture.

The announcement included the expansion of its Chicago cloud data center region at Centersquare’s Lisle, Illinois location. This features an AMD GPU supercompute cluster, powered by several thousand of AMD’s MI300X machine learning accelerators.

The company also extended its collaboration with AMD to make the new Instinct MI300X accelerators and ROCm open software available within Vultr’s composable cloud infrastructure.

It now has hundreds of thousands of active customers across 185 countries for its global cloud compute, cloud GPU, bare metal, and cloud storage solutions, the company said.

"We are excited to partner with Vultr to accelerate the growth of a standout leader in the cloud infrastructure industry," said Tanisha Keshava Bellur, Partner at LuminArx.

"Among the many management teams and companies we’ve evaluated in the AI infrastructure space, Vultr distinguishes itself through exceptional leadership, outstanding execution, and a proven track record of delivering high-performance cloud infrastructure for over a decade."

Researchers have exposed an issue with the memory implementation on AMD’s data center chips that could threaten the integrity of data, but the chipmaker has hit back at the claims.

In a paper due to be presented at IEEE in 2025, researchers from University of Lübeck, KU Leven, and University of Birmingham highlighted a potential weakness in AMD’s secure encrypted virtualization (SEV) technology.

Dubbed ‘badRAM’, the paper outlines how attackers could manipulate the SEV system to allow unauthorized access to encrypted memory on the processor.

In a post dedicated to explaining the badRAM attack, researchers explained how the SEV technology was intended to protect processor memory in virtual machine (VM) environments through encryption.

“AMD's Secure Encrypted Virtualization (SEV) is a cutting-edge technology that protects privacy and trust in cloud computing by encrypting a virtual machine's (VM's) memory and isolating it from advanced attackers, even those compromising critical infrastructure like the virtual machine manager or firmware.”

But the paper warned that if correctly exploited the threat actors could access data used by the microprocessor, and potentially read and even overwrite the encrypted content.

The researchers further detailed the underlying premise of the exploit, whereby attackers could use “rogue memory modules” to deliberately provide false information to the processor during startup.

Using a test rig that cost them just $10, kitted out with a Raspberry Pi Pico, and a DIMM socket to hold the RAM, the team was able to successfully exploit the flaw by fiddling with the serial presence detect (SPD) metadata to circumvent the SEV encryption.

“We found that tampering with the embedded SPD chip on commercial DRAM modules allows attackers to bypass SEV protections — including AMD’s latest SEV-SNP version,” the badRAM.eu website explains.

“For less than $10 in off-the-shelf equipment, we can trick the processor into allowing access to encrypted memory. We build on this BadRAM attack primitive to completely compromise the AMD SEV ecosystem, faking remote attestation reports and inserting backdoors into any SEV-protected VM.”

BadRAM flaw only medium severity due to high barrier to entry for attackers

In a security bulletin issued by AMD the chip giant outlined the issue, tracked as SB-3015 as follows.

“A team of researchers has reported to AMD that it may be possible to modify serial presence detect (SPD) metadata to make an attached memory module appear larger than it is, potentially allowing an attacker to overwrite physical memory.”

The CVE description described the issue as stemming from improper input validation for DIM SPD metadata that would allow an attacker with certain levels of access to potentially overwrite guest memory.

The issue was only classified as a medium severity threat warranting a 5.3 rating on the CVSS owing to the high level of access required by a potential attacker.

AMD said the issue is better described as a memory implementation issue, rather than an AMD product vulnerability, adding that the barriers to executing the attack are very high, explaining it being given a medium severity rating.

In a statement given to ITPro, AMD outlined the types of access that an attacker would need to exploit the issue, providing some mitigation strategies clients can take.

“AMD believes exploiting the disclosed vulnerability requires an attacker either having physical access to the system, operating system kernel access on a system with unlocked memory modules, or installing a customized, malicious BIOS,” AMD advised.

“AMD recommends utilizing memory modules that lock Serial Presence Detect (SPD), as well as following physical system security best practices. AMD has also released firmware updates to customers to mitigate the vulnerability.”

AMD has announced a 4% cut to its global workforce as part of a realignment of business priorities.

According to a statement sent to ITPro, AMD said it would be taking a number of steps ultimately resulting in a workforce reduction.

"As a part of aligning our resources with our largest growth opportunities, we are taking a number of targeted steps that will unfortunately result in reducing our global workforce by approximately 4%,” the statement said.

AMD did not state what those specific growth opportunities were, though it said it was committed to ensuring impacted employees are treated with respect and are assisted through the transition.

A previous Securities and Exchange Commission (SEC) filing stated that, as of December 2023, AMD employed a total of 26,000 staff members internationally. This means the layoffs could be equivalent to over 1,000 staff members.

The announcement comes after a promising charge in the AI space for AMD, with the firm’s Advancing AI event showing off strong footing in terms of the critical foundations needed for the technology.

At the firm’s Q2 2024 earnings call, AMD reported a 405% increase year-on-year in operating income in its data center unit. One analyst from IDC told ITPro that the firm is in a strong position to push hard on AI in 2025.

The firm’s Q3 earnings resulted in a 7% dip in shares according to reporting from Investopedia, though, with net income falling slightly below some estimates.

AMD isn’t the only chip giant making cuts

The move from AMD follows a raft of workforce cuts at industry rival, Intel, which announced plans to cut 15% of its workforce in August, equivalent to around 15,000 roles across its global workforce.

Speaking at the time, CEO Pat Gelsinger said the move was part of a cost structure realignment designed to change the firm’s operations, as it hadn’t been seeing enough revenue growth from technologies like AI.

Both AMD and Intel have been playing catch-up with industry rival Nvidia over the last two years. The chipmaker’s market value has surged due to soaring demand for its high-end chips, with big tech firms ramping up AI adoption and flocking to the firm.

Intel’s revenue had been falling before the August layoffs, and Gartner analyst Alan Priestly told ITPro the firm needs to focus on developing a clear product offering for AI in order to capitalize on the trend.

What do you think of when you imagine a supercomputer? Rows and rows of servers, certainly, but what about the facility it’s housed in? Whatever springs to mind, it’s probably not a papermill, yet that’s exactly where I went when I traveled to see Lumi, the fastest supercomputer in Europe.

Affectionately referred to as ‘the Queen of the North’, Lumi – which both stands for Large Unified Modern Infrastructure and is the Finnish word for snow – is tucked away on the outskirts of the city of Kajaani. The area, once known for its forestry and paper manufacturing, is now making a name for itself in IT, according to Pekka Manninen, director of science and technology at CSC, which runs both Lumi and the neighboring Finnish national supercomputers, Mahti and Puhti.

While Lumi is located in Finland and run by a company that’s 70% owned by the Finnish state and 30% by Finnish higher education institutions, it’s very much a European project. It’s part of EuroHPC, which also comprises Leonardo in Italy and Marenostrum 5 in Spain.

EuroHPC, which is a public-private partnership, also provides some of the funding for Lumi, along with the 11 European nations that make up the Lumi Consortium.

How big is big?

Physically, Lumi’s housing is imposing. It occupies most of a former machine hall, which is about two stories high, and roughly 50 meters long. In accordance with its namesake, Lumi was designed to evoke the idea of a snow bank: White, with small cut-outs that let light sparkle through from the data center inside and is lit from the outside in blue.

Inside, according to Top500, is the fifth most powerful supercomputer in the world. It consists of five aisles of HPE Cray Supercomputing EX4000 system cabinets, inside which are eight compute chassis, each with room for eight compute blades, eight HPE Slingshot switch blade slots, and one power/signal midplane.

Processing power for Lumi is provided by AMD and it contains almost 12,000 MI250x GPUs and 262,000 AMD Epyc Milan CPU cores. That doesn’t mean it’s operating as a single machine, though. There are eight partitions, each with a different specialty:

• LUMI-G: GPU partition, with 11,900 AMD Instinct MI250X GPUs
• LUMI-D: Data analytics, with graphics GPUs for analytics and data visualization
• LUMI-F: Accelerated storage. Using HPE Cray ClusterStor E1000 storage systems, it’s a 9PB flash-based storage layer with a read bandwidth of 3TB/s and extreme IOPS capability
• LUMI-P: 80PB parallel file system
• LUMI-O: Object storage with 30PB encrypted object storage for storing, sharing, and staging data
• LUMI-Q: Quantum processors
• LUMI-K: Container cloud service
• LUMI-C: A supplementary CPU partition with 262,000 AMD Epyc CPU cores

For customers that want to use more than one of these partitions, there’s also a high-speed interconnect using HPE Slingshot. This allows resources from several of the partitions to be used within a single run.

While Lumi’s outer casing is atmospherically lit, inside the room is bright and echoes with the noise of its servers and the infrastructure that keeps the supercomputer cool. It’s not completely deafening – although I picked up a pair of earplugs, others who were also on the tour chose not to and seemed to bear no ill effects.

It would probably be a different story if Lumi were air-cooled, but the direct water cooling and what happens to it once it has passed through the system is part of what makes Lumi special.

Running clean

Lumi is one of the greenest supercomputers. When it came online in June 2022, it immediately took the number three spot in the Green500 list of the most energy-efficient supercomputers. It has since dropped down the rankings and in June 2024 – the most recent rankings at the time of publication – was sitting at number 12.

But this doesn’t tell the whole story. While like any supercomputer Lumi requires a huge amount of energy to run, it’s all sourced from a local hydroelectric power plant that predates it. This means that from an energy use point of view at least, Lumi is carbon neutral.

Another growing concern about large data centers and supercomputers is the amount of water they use. Frontier, the world’s fastest supercomputer, requires 6,000 gallons of water to run through its system per minute to keep it cool, according to Bloomberg.

As my co-host and I recently discussed on the ITPro Podcast, this water must be fresh with as few impurities as possible. This is having a major impact on water resources in some towns and cities around the world that are close to large data centers, particularly in more arid regions. Additionally, the water that has passed through the data center – be that a cloud computing facility, supercomputer, or proprietary – is hot. While it varies from facility to facility, the average is approximately 40ºC, which is the case for Lumi.

Once again, Lumi’s unique geographic location means its impact is far lower than that of a system located in many other regions. Finland is known as the “land of a thousand lakes” and that’s not just a boast from the Finnish tourism board – in fact, it’s something of an understatement.

There are actually 187,000 lakes in Finland, which according to World Atlas is enough for one lake per 26 Finns. In short, there’s plenty of fresh water to go around, and having a supercomputer nearby isn’t going to impact the amount of potable water available to the residents of Kajaani.

Lumi even contributes to another part of the city’s infrastructure. The 40ºC water it produces is fed into a local district heating boiler, which is conveniently located on the same industrial site. There it’s heated up further and distributed to domestic buildings in the city for heating during the winter months. While this may be a novel idea for people living outside the Nordics, in Finland it’s the most common form of heating, according to industry body Finnish Energy.

“There are some houses that are still directly electrically heated, like private houses, but virtually all apartments and even most of the houses [in Kajaani] are district heated,” Manninen tells me.

This is certainly beneficial for a city where average temperatures are between 1ºC and -11ºC for six months of the year – even in July the mercury rarely hits 20ºC. Eventually, some of this water once it has cooled on its journey around the city comes back to Lumi and the process starts again.

But wait, there’s more.

Even the building that houses the facility itself is recycled. The last roll of paper to be produced at the paper mill that once called the building home came off the production line in 2008 and the machines themselves were dismantled two years later.

It was another nine years before Lumi was built, but the fact no new concrete was poured to accommodate it lends another stripe to its environmentally friendly credentials, as well as those of Mahti and Phuti.

Putting science at the core

So what kind of workloads is this facility running? In the words of Katja Mankinen, a data scientist at CSC, “tons”.

“In Lumi, you can build everything, from the smallest scales [such as] simulating matter [and] particles, how they interact, what are their properties, to what happens in the universe at the galaxy scale,” Mankinen told the handful of assembled journalists.

“We also have projects that help more societal issues – how to cure cancer, how to create personalized medicine, how to help people who have health problems,” she added.

One of the most high-profile permanent residents is Destination Earth, an ambitious project that seeks to create a digital twin of the planet Earth. Currently, according to scientists at CSC, Destination Earth can model climate and extreme weather events with unprecedented accuracy. The official timeline for Destination Earth also includes unspecified “enhancements” by 2026, with the ultimate aim of a “full digital replica of the Earth” by 2030.

Outside Destination Earth, Manninen tells me that a little over 50% of CPU cycles are used for training neural networks. This could include large language models (LLMs) for example or, increasingly, generative AI.

One way that this 50% is being used is by researchers from the University of Turku in the southwest of Finland. They have trained LLMs in minority languages, starting with Finnish before moving on to other Nordic languages, as well as now English and code.

Manninen expands on this: “Commercial LLMs are getting better, but they're still not perfect and there’s only a small section of the parameters that are actually dedicated to these kinds of marginal languages, like Finnish.

“These are important initiatives, like national LLM projects that take place as academic initiatives,” he adds, while also pointing out that AMD is involved in another project to create an LLM that knows all the official languages of the European Union.

The 50% not occupied by LLMs and generative AI is “pure high performance computing (HPC)”, Manninen says – floating point 64-bit workloads for physics, climate science, and so on.

“There’s also a niche – but a growing niche – in converged HPC and AI,” he says. “For example, there was a material science case where simulation data was being produced and then you train the [neural] network on the go as the actual production to replace simulations or speed up science. It’s still small, but it’s a growing percentage.”

Organizations wishing to use Lumi resources can apply through the official website, although there are stipulations around who can use it – it’s an EU initiative after all. However there are some benefits, not least that while organizations can keep the results of their research private if they choose to make them public then it’s “basically free to use”, says Mankinen. There’s also no need to move to Finland, let alone somewhere as distant as Kajaani, as everything can be done remotely.

The future

CSC’s future plans for Lumi are rather poetic, given its history as a factory.

Adjacent to the hall that holds Lumi is another almost identical room that once would have been filled with machinery for the paper mill but currently lies completely empty, save for a few wooden pallets and boxes. It seemed strange initially that we were shown this room on our tour, but as it turns out within the next few years it may be home to a new kind of factory.

Manninen told me that CSC hopes to host one of the EU’s AI Factories in this cavernous hall and that it is “well positioned” to do so. The call for expressions of interest went out on 10 September 2024 and continues until the end of the year.

It should be some time in 2025 that the team looking after Lumi find out whether it’s time to call in the builders. Whether or not she gets a next door neighbor, though, the Queen of the North is still an example of how a technology that could have serious negative environmental impacts can be run in a more environmentally friendly way. While its location is somewhat unique in terms of its geography, infrastructure, and climate, there are surely more universal lessons that can be taken into future supercomputer builds elsewhere in the world.

Intel and AMD have announced they will partner up to create an ‘x86 ecosystem advisory group’ in a move some experts think will help the pair fight off competition from UK chipmaker Arm.

The x86 is a complex instruction set computer (CISC) architecture that powers most of the world’s CPU market - according to Statista, Intel’s x86 processors made up 71% of laptop CPUs in Q3 2024.

Arm, meanwhile, produces a rival CPU architecture with the aim of cutting in on x86’s action in the industry.

Ben Bajarin, CEO and principal analyst at Creative Strategies, told ITPro that while the x86 architecture still dominates the CPU market, this move could help to fight off challenges from the firm.

The new advisory board will bring together several big names in tech as founding members, including Broadcom, Dell, Google, Hewlett Packard Enterprise (HPE), HP, Lenovo, Meta, Microsoft, Oracle, and Red Hat.

Interoperability is the key focus here, and the group will look to expand the x86 ecosystem through enhanced compatibility across different platforms, simplified software development, and provisions for developers.

The group wants to create a more unified set of instructions and architectural interfaces to ensure predictability and consistency across x86 products. To do this, the pair will field technical input from x86 hardware and software communities.

For x86 users, the intended benefits are greater levels of customer choice, standardized product interfaces across Intel and AMD x86 offerings, and greater levels of integration with operating systems, frameworks, and applications.

“We are on the cusp of one of the most significant shifts in the x86 architecture and ecosystem in decades – with new levels of customization, compatibility and scalability needed to meet current and future customer needs,” Intel CEO Pat Gelsinger said.

Will x86 stay on top?

At present, x86 commands a 90% share of the data center CPU market and 90% of the client CPU market, according to Bajarin. The advisory group is fighting off Arm in those areas more than anything else, he added, highlighting the necessity of the move.

Arm is a sizable competitor though. The firm has over 90% share of the mobile applications CPU market, according to Statista, and a 65% share of the global Internet of Things (IoT) market.

Bajarin said while the launch of the new advisory group could enable member firms to better shape the future of x86, there are still lingering concerns with regard to intel and AMD.

“This is a welcomed effort, but I'm not sure if it means more consistent and compatible extensions or not, as I expect Intel and AMD to still maintain some degree of differentiation with their versions,” Bajarin said.

There are a lot of underlying architectural characteristics Bajarin expects this move to have an impact on, though, and the deal will likely bolster more consistent software and hardware experiences.

Bajarin questioned how the move would help the x86 fit into the new world of AI, which AMD seemed to suggest it would. The chip maker emphasized the importance of a robust x86 ecosystem, adding that it will be crucial in the current landscape characterized in part by AI workloads.

Seeing as the x86 architecture doesn’t play much of a role outside of host CPUs in data centers, and that most AI workloads run on GPUs or non-x86 AI accelerators, Bajarin remains unconvinced about how the move would fit into plans for AI.

AMD chief executive Lisa Su ended her keynote speech at the company’s Advancing AI event with a comment marking her tenth year in the hot seat at the chip maker. It was a perfect week to celebrate such a major milestone.

The annual conference saw a raft of exciting new announcements for the firm, including launches for its new MI325X Instinct GPUs, EPYC CPUs, and Ryzen series to help AMD stamp its mark on the burgeoning Copilot+ PC market.

A sense of optimism permeated the keynote session and indeed the whole conference, with Su and fellow executives highlighting the company’s strong position as it moves into a critical period in the year ahead.

“We’re no longer a contender, we view ourselves as a leader,” Dan McNamara, SVP and GM at AMD’s server business unit, told ITPro in a briefing.

All told, AMD appears to be in a very good rhythm right now - and it’s been a long time coming. Under Su’s leadership, the chip maker has clearly been playing the long game: laying critical foundations, building out its ecosystem, acquiring wisely, and biding its time. And it’s beginning to pay dividends.

In 2017, AMD held a zero percent share of the data center market by revenue – today it stands at 34%, underlining the company’s rapid growth in this domain.

A glimpse at this surging potential for AMD came in its recent quarterly earnings call, which saw the company record a 405% increase in operating income in its data center unit compared to the same quarter in the year prior.

With the launch of its new Instinct GPU series, more promise awaits the chip maker. The MI325X - and future iterations roadmapped at the event - could be the ticket to tackling Nvidia’s dominance in the market given it’s more than a match for its current H200 Tensor Core GPU.

AMD told attendees that the MI325X can outperform the H200 by up to 40% in certain AI inference benchmarks, for example.

AMD is ready for major gains in 2025

Andrew Buss, senior research director for EMEA at IDC, tells ITPro that AMD appears to be in a prime position to launch a major push in the market in the year ahead and could gain even more ground on Intel.

“I think AMD has done very well in the last number of years since Lisa took over,” he says. “As we saw, they’ve gone from zero percent data center share - particularly for server CPUs, up to around a third of the value of the x86 server market. So they’re very potent compared to Intel.

“If you actually look at the progress of something like an ARM-based server, AMD has actually done far better,” Buss adds. “I think that shows that if you've got the right product with a performance advantage, a cost advantage, or efficiency advantage and you can deliver on your roadmaps and your product promises, you can be accepted.”

AMD did “a lot of the hard work around 2018 to 2020,” Buss notes, pointing to its ability to continually improve upon product releases across its main target areas as further proof of its sharpened focus.

“I think now we're seeing them expanding,” Buss says. “They’ve made some good acquisitions. They’ve always had the graphics division but maybe they hadn’t focused on AI as much as Nvidia had.

“Nvidia caught the generative AI wave, and AMD’s recognized that. Now with the Instinct series, the latest generation, the 300 series, is much more tuned to that.”

It must be frustrating for Nvidia’s competitors. In recent years, the firm’s meteoric rise has captured the attention of the media and tech industry stakeholders alike, with the generative AI race surging it to record profits and a formidable market cap.

While there are no concrete signs that the company will be knocked off its perch any time soon, this attention has enabled competitors such as AMD to innovate somewhat out of the limelight. This breathing space might be the best thing AMD could’ve hoped for.

Moving forward, the chip maker can frame itself as the best viable alternative to Nvidia. While exact details on pricing were left unmentioned, much of the messaging around its new generation chips pointed toward them being a more cost-effective option than Nvidia’s pricey hardware.

Moving forward, the chip maker can frame itself as the best viable alternative to Nvidia. While exact details on pricing were left unmentioned, much of the messaging around its new generation chips pointed toward them being a more cost-effective option than Nvidia’s pricey hardware.

AMD has announced a sweeping expansion of its high-performance networking portfolio in a bid to supercharge data center efficiency in the age of AI.

Officially unveiled at its Advancing AI conference in San Francisco this week, the chip maker hailed the launch of its new Pensando Salina data processing unit (DPU), which is aimed specifically at streamlining AI cluster performance.

AMD said the new DPU offers 2x performance, bandwidth, and scale compared to previous generations and can support 400G throughput. All told, this means faster data transfer rates and ultimately more efficient infrastructure capable of contending with surging AI workload demands.

The importance of networking has grown significantly since the onset of the generative AI race in late 2022 and has emerged as a key bottleneck to data center efficiency, AMD executives told attendees.

This is a double-whammy problem many organizations face, with efficiency on both the front and back end becoming a key focus.

On the front end, the efficient delivery of data to an AI cluster is critical, while performance on the back end, which manages data transfers between clusters, means communication between central processing units (CPUs) and graphics processing units (GPUs) within these clusters is vital.

Blockages within these networking paths can create serious problems for data center operators, resulting in lower performance and heightened costs.

The Salina DPU is designed to tackle these potential congestion issues within the front end. The second announcement from AMD focused on the back end with the launch of the Pensando Pollara 400 network interface card (NIC).

In a press briefing ahead of the event, AMD executives revealed this is the industry’s first Ultra Ethernet Consortium (UEC)-ready AI NIC which “reduces complexity of performance tuning and helps improve time to production”.

AMD’s networking focus shows maturity

Andrew Buss, senior research director for EMEA at IDC, told ITPro that while this announcement may not have hit the headlines to the extent of its Instinct GPU or AI PC reveals, it marks a significant moment for the chip maker.

“To be strong in infrastructure, you’ve already got to be strong across the board, because you want to be able to store the bits, move the bits, and change the bits,” he said.

“They've got a pretty good stack, and I think they're well set to drive that forward, particularly with Ultra Ethernet. So to me, some of the more important but maybe less visible announcements that they did make were this new DPU and the front-end and back-end in the network and the fabric and ultra Ethernet.”

Improving networking communication will be the key to optimizing large scale clusters, Buss added, and AMD appears very much focused on ramping up efforts in this domain.

Hints toward this strategy at AMD have emerged in recent months, especially when one examines recent acquisitions, such as the $4.9 billion deal for ZT Systems in August.

“I think that shows maturity,” he told ITPro.”AMD just recently acquired ZT systems as well. So what we're going to see is them following Nvidia's footsteps of being able to engineer systems-level approaches.”

AMD has unveiled its latest weapon in the battle to dominate the burgeoning Copilot+ PC market with the launch of the Ryzen AI Pro 300 series.

Specifically catered toward enterprise users, the move by AMD looks to arm businesses with more powerful AI chips to bolster on-device AI and security capabilities.

The new series comes in three iterations, the AMD Ryzen AI 9 HX Pro 375, AMD Ryzen AI 9 HX Pro 370, and AMD Ryzen AI 7 Pro 360.

Officially announced at the chip maker’s Advancing AI conference in San Francisco, the new series marks a notable increase in capability and performance compared to the previous generation and the competition.

Detailing benchmarks, the company said the top-of-stack HX Pro 375 offers up to 40% higher performance and 14% faster productivity performance compared to Intel’s Core Ultra 7 165U.

Moreover, the addition of a more powerful NPU represents a key differentiator, according to AMD, boasting a “cutting-edge” 50+ NPU TOPS of AI processing power.

Notably, given it’s the only X86 processor on the market with the required TOPS to meet Microsoft’s Copilot+ threshold, the move could mark a significant moment for the firm and give it a competitive edge.

AMD capitalizing on the AI PC craze

The move from AMD comes amid a period of growing interest in AI PCs, with a host of major manufacturers ramping up efforts to roll up AI-powered devices in both the consumer and enterprise markets.

Intel has moved aggressively in this domain across 2024, having identified this as a key potential growth opportunity late last year. Intel CEO Pat Gelsinger went so far as to claim the AI PC would be the “star of the show” in 2024.

But AMD has been gaining ground, particularly in the notebook domain. Analysis from Mercury Research earlier this year noted the chip maker gained “significant server, desktop, and notebook revenue share year on year”.

A key factor in this was heightened demand for its 4th Gen EPYC and Ryzen 7000 series processors, according to Mercury Research.

“Enterprises are increasingly demanding more compute power and efficiency to drive their everyday tasks and most taxing workloads,” said Jack Huynh, SVP and general manager for AMD’s computing and graphics group.

“Our this generation AI-enabled processors for business PCs deliver unprecedented AI processing capabilities with incredible battery life and seamless compatibility for the applications users depend on.”

AMD outlined ambitious targets for the new Ryzen range, noting that it intends to ship more than 100 different notebook models with OEM partners this year, including via Lenovo and HP.

With high stakes and potential battle between AMD and Intel brewing in this domain, IDC senior research director for EMEA, Andrew Buss, told ITPro the latest announcement shows the race is very much on. Both Intel and AMD have progressively bolstered capabilities recently and we can expect this battle of attrition to accelerate.

“If you look at how AMD has expanded its NPU over the last two generations, moving from the [Ryzen] 7000 and 8000 series into Strix point, it’s a much bigger NPU,” he said.

“And if you look at Intel Meteor Lake versus Luna Lake, Luna Lake is a much bigger NPU as well. So it’s becoming much more important.”

Moving forward, Buss suggested this process of upgrades will essentially become the new normal within the market, and devices not offering these capabilities will eventually be weeded out. All told, the days of non-AI PCs could be coming to an end - or at least forming a distinctly separate market.

“This is the natural evolution, just how pretty much every laptop now has an iGPU in it, every laptop in three years will have an NPU of various strengths,” he told ITPro.

“We've got existing designs and markets that are pre AI PC, they'll have to be managed out the market as a long tail, so they’ll naturally fall away and become uncompetitive designs and move from high end, to the mid-range, to the low end and help the market.

AMD has announced the launch of its new Instinct MI325x accelerators as competition heats up with Nvidia in the AI space.

Unveiled at the chip maker’s Advancing AI conference in San Francisco, AMD said the new range is designed specifically to drive performance and efficiency in contending with “demanding AI tasks”.

The new Instinct range is aimed at supercharging foundation model training, fine-tuning, and inference, the company said.

The announcement from AMD marks the latest in a slew of progressive upgrades to its Instinct graphics processing unit (GPU) series in recent years and forms a key focus in its ongoing roadmap to drive AI infrastructure advances for hyperscale customers.

In a press briefing ahead of the Advancing AI event today, AMD executives waxed lyrical about the latest iteration, emphasizing performance capabilities compared to competitors such as Intel and Nvidia.

But it’s the latter of these that enterprises will be keen to see the comparable benchmarks for.

AMD said the MI325X accelerators provide “industry-leading” memory capacity and bandwidth, boasting 256GB of HBM3E supporting 6TB/s. All told, this equates to 1.8x more capacity and 1.3x more bandwidth than Nvidia’s H200 series.

In terms of LLM inference capabilities, the company told assembled press the MI325X can provide up to 1.3x the inference performance on the Mistral 7B model, for example, along with 1.2x inference performance on the Llama 3.1 70B range.

AMD revealed the MI325X accelerators are currently on track for production shipments in Q4 2024, and are expected to have widespread availability from a host of platform providers by Q1 2025.

These include Dell Technologies, HPE, Lenovo, Supermicro, and more.

AMD’s roadmap is accelerating

Looking ahead, AMD execs gave attendees a glimpse at what’s in store with regard to future offerings in 2025 and beyond. The chip maker previews the next-generation AMD Instinct MI350 series accelerators at the annual conference, highlighting its rapid-fire efforts in the next few months.

Based on the AMD CDNA 4 architecture, the MI350 series accelerators are designed to deliver a whopping 35x improvement in inference performance compared to CDNA 3-based accelerators.

This will also include increased memory capacity, the company confirmed, reaching 288GB of HBM3E memory per accelerator.

Development of the MI350 series is on track, AMD said, and will be available during the second half of 2025.

Welcome to ITPro's live coverage of the AMD Advancing AI conference in San Francisco.

It's a chilly morning here in San Francisco, but we're up and rearing to go ahead of the conference, which is due to kick off at 9am PST.

We've got a jam-packed agenda ahead of us at the Moscone Center today, and yesterday saw assembled press get a sneak peak at all the news and announcements we can expect to see in the opening keynote session – and we can assure you there's going to be a lot of talking points.

In the opening keynote we'll hear from a variety of executives at AMD, including chief executive Lisa Su.

Make sure to keep tabs on our rolling live coverage here throughout the day.

It's clear that AMD has been cooking up a storm so far in 2024 as it primes itself for a battle with Nvidia.

All eyes have been on the latter over the last two years since the emergence of generative AI, but AMD has been putting in the groundwork of late to catch up with its great rival.

Today we'll hear more about how it plans to ramp things up in the year ahead, but we have seen a raft of exciting acquisitions, including two in the last couple months alone. These included a $4.9 billion deal to acquire ZT Systems, which marked a huge statement of intent from the chip maker.

In July, AMD also announced the $655 million acquisition of SIlo AI amidst its big AI push.

We've still got a little while until the opening keynote session kicks off, but in the meantime why not check out our preview ahead of the conference for a taste of what we can expect to hear about?

• AMD has put in the groundwork for a major AI push while the tech industry fawned over Nvidia

It's still early - and we are expecting a torrent of people to converge on the Moscone Center this morning - but I have to say the city center in San Francisco has been a breath of fresh air compared to the carnage that Dreamforce brought with it just a couple of weeks prior.

No sea of lanyards or the cacophony of car horns in chaotic traffic.

We're expecting to hear about a big push on AI PCs from AMD today. The rise of the AI PC has been a big talking point over the last year, with manufacturers and chip makers alike ramping up development in this space.

Intel CEO Pat Gelsinger claimed last year that the AI PC would be the "star of the show" in 2024, and he wasn't wrong. This has been a big area of focus for Intel, one of AMD's key competitors in this space.

But what's AMD doing in this regard? Its Ryzen series is a key focus for the company in its big AI PC push, so we'll likely hear more on this during the opening keynote.

There has been some concern from industry stakeholders on the Ryzen series, however.

After AMD unveiled its Ryzen Pro enterprise-focused processors, the company promised significant performance gains, but reports in April suggested they were unlikely to meet the steep performance threshold required to be designated as an 'AI PC' by Microsoft.

A key talking point here was that the AI processing power of these chips appeared to fall short of the 40 trillion operations per second (TOPS) standard set by Microsoft.

You can read more on this in our coverage below.

• AMD's new series of AI-enabled processors likely fall short of Microsoft’s performance requirements for next-gen AI PCs

With this in mind, we could see AMD come out guns blazing with a new update to the Ryzen series.

Like we said, a real contrast in the volume of people here at the Moscone Center today.

We are seated now and good to go. Just a matter of time until we hear from Lisa Su in her opening remarks here at AMD Advancing AI. A good buzz around the keynote hall ahead of things commencing.

And here we go! AMD chief executive Lisa Su is on stage to kick things off at Advancing AI.

"Let's talk a little bit about AI," Su says. "Actually, we're going to talk a lot about it."

Setting the tone here right off the mark.

We're getting a brief rundown of AMD's current strategy.

"It's about employing the entire ecosystem - the cloud, OEMs...our goal is to create an open ecosystem," Su says.

Su says AMD is "really committed" to driving open innovation in high performance computing and AI infrastructure. This has been a recurring talking point in the company's strategy of late - ecosystem building and collaboration.

"When we think about AI, it really is about using the right compute for the right application," Su says.

There's really no 'one-size-fits-all' approach to AI, and AMD is keen to enable enterprises to get those creative juices flowing and adopting the technology on their terms, at their pace, and based on their unique individual needs.

We've moved onto the data center portfolio here. A really exciting area for AMD right now - big gains and big revenues boosts in recent months.

AMD EPYC now boasts a record market share - and it's growing, Su says. At the end of the second quarter, AMD exited with a 34% revenue share in this domain.

We've got our first partner on stage now from Google Cloud.

"The demand for AI compute power is insatiable."

So how's this benefitting Google Cloud? Well, leveraging EPYC machines, the tech giant has cut costs by 40% and delivered a 15% performance improvement.

Big numbers here and when you look at this from a financial perspective, very significant for the hyperscaler.

Data center AI demand is absolutely skyrocketing. If you're a regular reader of ITPro then you'll have come across some of the various bits of research we've covered on this subject.

Insatiable demand from enterprises globally has prompted a massive surge. This, Su says, represents a massive opportunity for AMD to capitalize on this trend with its Instinct GPU series.

Looking ahead, the numbers involved here are simply mind boggling.

So what's AMD's big selling point here then with regard to GPU capabilities?

The Instinct MI300X series is the chip maker's flagship product here, and in terms of performance it's far exceeding what competitors can provide.

The MI300X outperforms Nvidia's H100 series on several key benchmarks, Su explains. All told, this represents 1.3x higher performance on Llama 3 and Mistral-based workloads.

We now have another partner joining Su up on stage. This time it's Karan Batta, SVP at Oracle.

AI, naturally, is the biggest talking point here. Oracle is seeing "incredible levels" of performance and efficiency on AMD GPUs, Batta says.

The company is continuing to scale up capacity for MI300X series GPUs for customers, Batta says. There's more to come in the year ahead as well.

Rapid fire customer chats here with Naveen Rao, VP of AI at Databricks joins Su on stage now to discuss the company's partnership with AMD.

"We've been on this journey for a while," Rao says. "We've achieved remarkable results."

Leveraging the MI300X the company has delivered marked compute improvements. It's a real "powerhouse" of a GPU, Rao says.

Satya Nadella, Microsoft CEO, joins Su now…just not on stage. We’re getting a run through a video chat with the duo discussing the pace of change and development in the AI space.

“It’s been incredible, the amount of innovation in the industry,” Su says.

AMD and Microsoft have deep, deep ties. Microsoft is creating an entirely new PC category with AMD at the moment with AI PCs.

In the last four years, Microsoft has been tapping into AMD’s AI innovation to support its own cloud innovation. It’s a very beneficial feedback loop for both organizations that’s paying dividends.

Microsoft is using the MI300 series which is delivering great benefits for key services such as its raft of various AI solutions and tools, as well as Azure.

“What we have to deliver is performance, per dollar, per watt,” Nadella says. A big focus on energy efficiency and performance to deliver cost benefits for enterprises.

That’s the one metric that matters, Nadella says. Given the current costs associated with AI development, unlocking these improvements is vital.

AMD really is pulling out the big hitters in terms of brand names here. Oracle, Microsoft, and now Meta.

Su is joined on stage now by Kevin Salvadori, VP for Infrastructure and Engineering at Meta.

Meta's sharpened focus on generative AI over the last two years has been underpinned by AMD Instinct GPUs and EPYC CPUs.

Salvadori admits that Meta is a demanding customer, but AMD has been on hand to match those expectations amid the company's big AI boom.

"We've deployed over one and a half million CPUs," Salvadori says. Huge numbers and serious scale.

Using the MI300x series in production has been "instrumental" to helping Meta scale its AI infrastructure and power inference. This has been crucial to support development of the Llama large language model.

While hardware is important, software is an equally crucial aspect of AMD’s current roadmap and strategy.

AMD platforms are powering some of the most powerful workloads on the planet.

We’re getting a run down on ROCm now, AMD’s open software stack that complements its hardware offerings.

“ROCm really delivers for AI developers.”

Part of this focus includes close collaboration with the open source community for AMD. The chip maker works closely with PyTorch, for example, as well as Hugging Face and vLLM.

With Hugging Face in particular, more than one million models run on AMD.

We've got a panel session coming up now with a host of industry stakeholders, including:

• Dani Yogatama, CEO at Reka AI

• Dmytro Dzhulgakov, CTO at Fireworks AI

• Ashish Vaswani, CEO of Essential AI

• Amit Jain, CEO at Luma AI

So how is ROCm helping support development at these trailblazing companies?

The open source approach that AMD employs has been highly beneficial to all four organizations, that much is clear. Furthermore, AMD GPUs are underpinning the development of the models currently in development at these firms.

"ROCm's open source platform really allowed us to move fast," Yogatama says.

Dzhulgakov at Fireworks AI says using the MI300 series has unlocked marked benefits for the company during its development process.

"This basically allows us to deliver really efficient deployment of AI models," he says.

Forrest Norrod, EVP and GM of the AMD data center solutions business unit takes to the stage now to give us a run through all the goings on in the firm's data center segment.

This aspect of the business has been booming of late, so a lot to be excited about from AMD's perspective.

It's safe to say AI has changed the game with data center operations, both on the front end and back end. Networking is also evolving at pace to contend with this new level of scale, Norrod says.

"Networks now must scale efficiently," he says. Poor networking can create major bottlenecks for AI clusters. "AI networking must match the unique challenging characteristics of AI workloads."

Ethernet has traditionally been the answer to this. "Quite simply, from cloud to enterprise, ethernet is preffered," Norrod says.

But while it's a clear winner in terms of scale and TCO, there are difficulties at play here and it does need to evolve.

General purpose networks can typically operate at under 50% utilization, Norrod says. But AI workload networks are operating at 95% utilization. This underscores the demands placed on network infrastructure by AI.

We’ve got another customer up with Forrest now, this time from Supermicro.

Vik Malyala, SVP for technology and AI at Supermicro says AMD has enabled the firm to create ‘kick ass’ products for customers.

“BY working closely with AMD, one of the things we do is further stretch performance limits,” Malyala says. The company uses EPYC CPUs, and Malyala says this has delivered marked improvements to both efficiency, but also in providing benefits for customers.

That's the hardware side of things though, Malyala says. the software aspect, like we heard earlier, is equally compelling and is highly beneficial.

AMD's open ecosystem has been critical, in this regard, Malyala adds.

AMD and Lenovo boast close ties, Norrod says, bringing Vlad Rozanovich, SVP at Lenovo's infrastructure solutions group.

"Over the last 15 year, we've actually grown our AMD footprint in that tier 2 cloud service provider by around 50%," Rozanovich says.

Over the last three years they've grown their AMD footprint three times over. Lenovo is well and truly all-in on AMD.

This is helping Lenovo really simplify AI for enterprise customers, Rozanovich says, noting that reliability and security are two areas where Lenovo really differentiates itself. A big talking point during the AI boom, and AMD is helping underpin this reputation, Rozanovich says.

"Lenovo and AMD are really driving AI to provide the best customer experience possible," Rozanovich says.

Lisa Su is back on stage now to round things off. We've heard a lot about data center infrastructure, but now we're going to be looking at AI PCs.

This is an exciting focus area for AMD and there's huge potential in AI PCs both in the consumer and enterprise markets. AMD has invested heavily in NPUs, Su says, and has been innovating aggressively in this space with its Ryzen series.

"When you look at our platforms, they're really built to deliver the best performance, multi-day battery life, security, reliability, all the things we need in enterprises," Su says.

Wouldn't be an AI PC stream without a mention of Microsoft - one of the leaders in this space with its Copilot+ range for Windows.

With that in mind, Su welcomes the final guest to the stage today, Pavan Davaluri, corporate VP for Windows and devices at Microsoft.

With AMD, Copilot+ PCs are going to hit new heights in terms of performance, Davaluri says.

"Speed matters, and Copilot+ PCs are fast," he says. AMD's Ryzen series plays a key role to help Microsoft meet expectations in this regard.

Everything we learned at AMD Advancing AI

And with that the keynote session at AMD Advancing AI is done and dusted. There was a lot to digest there but what was clear is that AMD has a concrete plan of attack to take on Nvidia in the year ahead and beyond.

In a keynote session filled with customer success stories and bold claims, there were a few big talking points with regard to product launches, so here's a rundown of everything we heard:

AMD Instinct GPU updates

The next iteration of AMD's Instinct GPU series, the MI325X, was officially unveiled.

According to AMD, the MI325X accelerators will "deliver leadership AI performance and memory capabilities" for even the most demanding AI workloads. Better still, they're more than a match for Nvidia's upcoming - and hotly-anticipated - Blackwell chip.

That's not all we heard on the Instinct front, however. Lisa Su gave us a glimpse at the roadmap for future releases, including the MI350 series, which is expected to begin shipping in the second half of 2025.

We even had a glimpse at the MI400 series, which we can expect to see rolling out in 2026.

DPU and NIC updates

We had an update with regard to AMD's DPU and NIC offerings, including the launch of the Pensando Salina DPU alongside the Pendanso Pollara 400.

Both focused on streamlining and maximizing AI infrastructure performance, these will help enterprise customers "optimize data pipelines," according to AMD and facilitate better GPU communication for high-performance AI systems.

Ryzen processors are getting a big boost

Toward the end of the keynote session, Lisa Su unveiled the launch of the Ryzen AI Pro 300 series processors.

These will feature AMD 'Zen 5' and XDNA 2 architectures and are aimed specifically at delivering better performance, battery life, and security for Copilot+ laptops used in the enterprise.

We've heard a lot about AI PC's over the last year, and AMD certainly seems bullish on this front for the year ahead.

EPYC processors target big efficiency gains

We've got new EPYC processors on the way, mainly the 9005 Series.

Again built on AMD's 'Zen 5' architecture, these new CPUs appear to be a source of great excitement at AMD.

These will help deliver "record-breaking performance" and energy efficiency, according to the chip maker. There's already been some initial adoption, as well, with Dell, HPE, Lenovo, an Supermicro all reporting positive results so far.

Advanced Micro Devices (AMD) might rank among the old guard of the global technology industry, but it’s by no means out of ideas or inspiration – and the coming years could see it kick things up a notch.

The widespread public emergence of generative AI in late 2022 flipped the industry on its head and acted as a springboard for success for several major companies, including Microsoft, OpenAI, and most of all Nvidia.

The meteoric rise of Nvidia over the last two years has been nothing short of remarkable. Much of the generative AI ‘boom’ has been underpinned by Nvidia’s graphics processing units (GPUs), resulting in record profits, a surging market cap, and companies scrambling to get their hands on its hardware.

But for long-term rival AMD, Nvidia’s time in the limelight may have provided exactly the breathing room they needed. AMD hasn’t exactly been wallowing in a pit of despair over the flashy headlines or journalists fawning over the founder’s fashion sense. If anything, it’s been priming itself for a battle.

AMD has been a hive of activity so far in 2024 and there is a prevailing sense that it’s gathering momentum. Acquisitions in particular have been a key point of interest here, with the company having invested more than $1 billion between August 2023 and 2024 to expand the AMD AI ecosystem and bolster its software capabilities.

July this year saw the firm acquire Silo AI as part of a $665 million deal. This marked a significant moment for AMD, enabling it to greatly expand its portfolio of end-to-end AI solutions. Just one month later, we saw another marquee deal and a serious signal of intent for the company with the acquisition of ZT Systems for a whopping $4.9 billion.

ZT Systems provides AI and compute infrastructure for a host of major hyperscaler cloud providers. In bringing the company under its umbrella, AMD is taking another major step toward expanding both its AI chips and hardware business.

Commenting on the deal at the time, AMD chair and CEO Lisa Su said the acquisition will enable the company to “significantly strengthen” its data center AI systems and “customer enablement capabilities”.

“This acquisition also builds on the investments we have made to accelerate our AI hardware and software roadmaps," Su added.

Alongside other deals – including AMD’s acquisition of open source AI software firm Nod.ai in October 2023 – Su’s ambitious ‘roadmap’ is forming, as AMD cooks up a diverse range of offerings for customers akin to Nvidia.

AMD’s investments show it building a comprehensive ecosystem focused around its ‘open standards’ approach to AI. So far, this strategy appears to be paying dividends for the company. In its Q2 2024 earnings report, published in late July, we saw concerning results in one segment of the business but with a notable diamond in the rough.

The firm’s revenue for the second quarter totaled $5.835 billion, marking a 9% year-over-year increase and a 7% boost quarter over quarter. Even as it marked a sizeable drop of 59% in its gaming hardware segment across the quarter – perhaps a discussion for another publication – AMD posted record sales of data center products and markedly increased sales of processors for client PC devices.

Data center hardware sales reached $2.834 billion in Q2 2024, up 115% year over year while its data center business unit recorded $743 million in operating income. That’s 405% higher than in the same quarter last year.

Overall, revenue for AMD’s data center segment surged by 21% compared to the previous quarter, underlining a period of rapid growth for the company.

Underpinning this impressive growth has been a sharp increase in shipments of both AMD Instinct GPUs and strong sales of its 4th generation AMD EPYC processors, both of which are proving highly popular for enterprises ramping up AI development due to their performance and efficiency.

AMD will be keen to capitalize on this performance moving into the closing months of the year and set itself up to challenge in 2025 onward. This week also marks an opportunity to set the stage and highlight this optimist at its Advancing AI event in San Francisco.

ITPro will be live on the ground in San Francisco covering the event, and we can expect to see a bullish stance from Su and AMD as at large as the company prepares for a fresh raft of product and service announcements.

AI PCs are set to make up the majority of all computers sold by 2027, according to new research, with adoption rates expected to surge in the coming years.

The PC market has largely stalled in recent years, with manufacturers experiencing a severe drought in consumer and enterprise demand, though that looks set to change.

Research earlier this year predicted a slight increase in sales thanks to a looming Windows refresh as Windows 10 hits end of support next year as well as the rise of AI PCs.

Now, research by AltIndex.com using data from Statista and Canalys shows that six-in-ten of all PCs sold by 2027 will be AI PCs. That's three-times more than are expected to be sold this year.

This uptick will be driven by manufacturers pushing AI PCs as their top-of-the-line products, the study noted, as well as their utility in industries relying on edge computing.

"Many consumers and businesses also turn to AI PCs for enhanced productivity and to handle tasks like machine learning, data analysis, and creative work more efficiently," a post from AltIndex.com added.

"The rise of generative AI applications has also fueled demand for these devices. With more and more users searching for an AI-enhanced PC, their annual shipments will skyrocket in the following years."

The data suggested AI PCs would make up a fifth of total PC sales this year, tripling to 60% by 2027 and topping 205 million by 2028.

What is an AI PC and why are they all the rage?

While there's debate, the simple definition of an AI PC is one that features dedicated hardware to run AI systems.

Microsoft defines it as a PC running a neural processing unit (NPU), which imitates the brain's neural network structure to run AI applications and tasks more efficiently, while leaving your CPU and GPU to handle the rest of your computing.

AltIndex points to Lenovo's Yoga Book 9i — released last year — as the first "AI-enhanced PC", but notes that AI accelerators and other hardware have since been integrated into the designs of many more PC manufacturers.

To be clear, most laptops will be perfectly fine running the current generation of AI, such as using AI-based software or tools, or solutions like Microsoft Copilot — you don't need anything special to access ChatGPT, for example.

Indeed, Microsoft doesn't specify any hardware requirements to run Copilot 365 at the moment, though there are licensing and software that must be in place first.

So why is Microsoft offering PCs designed for AI? To run more advanced Copilot+ PCs "experiences" — which includes controversial Windows Recall, for example — a computer needs 256GB storage, 16GB RAM, and a compatible processor or system on a chip.

So far, this only includes the Snapdragon X Plus or the Snapdragon X Elite. Other AI PC chips could include AMD's Ryzen AI 300 or Intel's Lunar Lake while Apple is working on its own in-house hardware, according to AlphaSense.

"The Copilot+ PC features are more than just apps," Microsoft says on its website. "They are advanced AI capabilities that require a Neural Processing Unit (NPU) with 40+ TOPS to work as intended. Without an NPU, these features, such as Paint Cocreator or all Windows Studio Effects, may not function properly or at all."

This all suggests that, going forward, such specs are what will become the minimum for running AI tools — and that means, assuming AI takes off, that ‘AI PCs’ will soon simply become PCs.

Girding its loins for a widely-anticipated battle with Nvidia, AMD has acquired ZT Systems for $4.9 billion.

ZT Systems, which provides AI and general purpose compute infrastructure for major hyperscale providers, will, AMD hopes, help expand its AI chips and hardware business.

The cash and stock deal includes a contingent payment of up to $400 million based on certain milestones being hit after the deal has closed - expected to be in the first half of next year. ZT Systems' data center infrastructure manufacturing business will be split off and sold.

"ZT adds world-class systems design and rack-scale solutions expertise that will significantly strengthen our data center AI systems and customer enablement capabilities. This acquisition also builds on the investments we have made to accelerate our AI hardware and software roadmaps," said AMD chair and CEO Lisa Su.

"Combining our high-performance Instinct AI accelerator, EPYC CPU, and networking product portfolios with ZT Systems’ industry-leading data center systems expertise will enable AMD to deliver end-to-end data center AI infrastructure at scale with our ecosystem of OEM and ODM partners."

Once the deal closes, ZT Systems will join the AMD Data Center Solutions Business Group. Chief executive Frank Zhang will lead the manufacturing business while ZT president Doug Huang will front the design and customer enablement teams.

Both will report to AMD executive vice president and general manager Forrest Norrod.

"We are excited to join AMD and together play an even larger role designing the AI infrastructure that is defining the future of computing," Zhang said.

"For almost 30 years we have evolved our business to become a leading provider of critical computing and storage infrastructure for the world’s largest cloud companies. AMD shares our vision for the important role our technology and our people play designing and building the computing infrastructure powering the largest data centers in the world."

AMD has been on an acquisition spree

AMD is the only serious challenger to Nvidia's dominance in the chip market, and hopes that this new acquisition will help it catch up.

The deal marks the latest in a series of investments by AMD aimed at strengthening its AI capabilities. In the last 12 months, it's invested more than $1 billion to expand the AMD AI ecosystem and strengthen its AI software capabilities.

In April, the company announced new processors for AI-enabled laptops and desktops. Similarly, in July, it agreed to buy Silo AI for $665 million to boost its portfolio of end-to-end AI solutions.

While AMD still lags behind rival Nvidia, the move could be good for competition in the long run, according to Rodrigo Liang, CEO and co-founder at SambaNova Systems. Liang noted that the acquisition could improve enterprise choice.

"AMD's acquisition of ZT Systems recognizes the growing demand for alternative options to Nvidia’s AI platforms," he said.

"Enterprises don’t want to be locked into a single chip vendor, they want the choice to implement the best full stack AI systems that solve their AI challenges."

Advanced Micro Devices (AMD) is a multinational semiconductor company renowned for its innovative computing and graphics products. Competing directly with giants like Intel and NVIDIA, AMD has carved out a significant place in the technology industry with its high-performance processors and graphics cards.

Here we have a comprehensive overview of AMD, including its history, product offerings, key figures, and what customers can expect from doing business with it.

A brief history of AMD

AMD was founded in 1969 by Jerry Sanders and a group of engineers from Fairchild Semiconductor. Initially, AMD focused on producing logic chips. In 1975, it entered the microprocessor market, which would become its core area of focus. The company grew steadily, facing various challenges and competitions, most notably with Intel.

In the early 2000s, AMD gained significant market share with its Athlon and Opteron processors. However, the company struggled during the late 2000s and early 2010s, falling behind Intel in performance. The turnaround came in the mid-2010s with the introduction of the Ryzen processors, which brought it back into strong competition with Intel.

AMD's strategic acquisitions, such as the purchase of ATI Technologies in 2006 and Xilinx in 2022, further expanded its product portfolio and technological capabilities, particularly in graphics and adaptive computing.

What does AMD sell?

AMD offers a diverse array of products that cater to a wide range of markets, from consumer electronics to data centers, gaming, and embedded systems. At the core of AMD's offerings are its microprocessors, which have become increasingly competitive in recent years. The Ryzen series, including Ryzen 3, 5, 7, and 9, targets consumers and enthusiasts with high-performance processors for desktop and laptop computers.

For business users, it provides the Ryzen PRO line, which incorporates enhanced security features and manageability. Budget-conscious consumers can turn to the Athlon series for everyday computing needs. In the server and enterprise space, AMD's EPYC processors have gained significant traction, offering high core counts and performance for data centers and complex applications.

Graphics processing is another key area for AMD, with its Radeon line of GPUs serving gamers and content creators. These range from entry-level to high-end cards, competing directly with NVIDIA in the consumer space. For professionals in fields like design, engineering, and media production, AMD offers the Radeon Pro series, delivering workstation-grade graphics solutions.

AMD also produces chipsets, which work in tandem with their processors to manage data flow in both desktop and laptop motherboards. In the embedded systems market, AMD provides specialised processors that find applications in industries such as automotive, aerospace, and telecommunications, as well as in Internet of Things (IoT) devices.

The acquisition of Xilinx in 2022 expanded AMD's portfolio to include field-programmable gate arrays (FPGAs). This move brought the Virtex and Zynq series under AMD's umbrella, offering high-performance FPGAs and system-on-chip solutions that combine FPGA fabric with ARM processor cores.

It also sells Accelerated Processing Units (APUs), which integrate CPU and GPU capabilities on a single chip. These are particularly appealing for systems requiring good graphics performance without the need for a discrete GPU. The company has also made significant inroads in the gaming console market with its semi-custom solutions, powering popular platforms like PlayStation and Xbox.

For data centers and high-performance computing applications, AMD offers the Instinct series of GPUs, designed to accelerate AI, machine learning, and complex computational tasks. This line directly competes with NVIDIA's data center GPU offerings.

AMD's mergers and acquisitions

Over the years, AMD has strategically acquired several companies to enhance its market position and technological prowess. In 2006, AMD made a pivotal move by acquiring ATI Technologies for $5.4 billion. This acquisition marked AMD's significant entry into the graphics market, allowing it to compete directly with NVIDIA. The integration of ATI’s advanced graphics technology enabled AMD to develop its renowned APUs, which combine CPU and GPU capabilities on a single chip, revolutionising the way computing and graphics processing were handled together.

In 2012, the company acquired SeaMicro for approximately $334 million to bolster its data center solutions. SeaMicro's innovative energy-efficient server technology was a perfect fit for AMD’s strategy to reduce power consumption while increasing performance in data centers. This acquisition allowed AMD to provide more competitive solutions in the burgeoning cloud and data center markets, further solidifying its presence in this critical sector of the tech industry​.

The acquisition of Xilinx in 2022 for $35 billion was another game-changer for AMD. Xilinx's expertise in Field-Programmable Gate Arrays (FPGAs) and adaptive System-on-Chips (SoCs) significantly broadened AMD's product offerings, particularly in adaptive computing. This acquisition not only strengthened AMD’s position in data centers but also expanded its reach into automotive and industrial applications, areas that are increasingly reliant on sophisticated computing solutions.

In August 2024, AMD completed the acquisition of Finnish firm Silo AI for $665 million. Silo AI, based in Helsinki, specializes in providing end-to-end AI solutions for enterprise organizations. Its open-source large language models (LLMs), including Poro and Viking — both multilingual — are built on its proprietary SiloGen model and optimized for AMD's platform.

Following the acquisition, Peter Sarlin, Silo AI’s co-founder and CEO, has taken on a leadership role within AMD’s Artificial Intelligence Group, reporting directly to Vamsi Boppana, AMD’s senior vice president. This acquisition reinforces AMD’s commitment to advancing AI capabilities, particularly for enterprise-level applications and multilingual AI solutions.

Also in August 2024, AMD announced it planned to acquire ZT Systems, a leading provider of AI infrastructure for hyperscale computing companies, in a deal valued at $4.9 billion. This acquisition was expected to significantly expand AMD's data center AI systems capabilities. The transaction was completed at the end of March 2025.

Key figures at AMD

Dr. Lisa Su, as Chair and CEO, is widely recognised for revitalizing AMD, driving it back to profitability and significant growth. Overseeing the company’s strategic direction is Victor Peng, the company’s president. Mark Papermaster, the CTO, is at the helm of AMD's technology and product development.

Supporting this leadership team are key executives like Jean Hu, the CFO, Darren Grasby, executive vice president and president of EMEA, and Forrest Norrod, executive vice president and general manager of AMD’s data center solutions. This cohesive leadership team has been instrumental in AMD's resurgence.

What can customers expect from doing business with AMD?

For IT decision-makers, partnering with AMD offers a compelling value proposition that addresses key enterprise concerns. AMD's product portfolio delivers high-performance computing solutions at competitive price points, particularly relevant for data center deployments, high-performance computing (HPC) environments, and virtualization infrastructures.

AMD's EPYC server processors have gained significant traction in the enterprise space, offering a strong performance-per-watt ratio that can lead to lower total cost of ownership (TCO) for data centers. This efficiency translates to potential reductions in power consumption and cooling costs, critical factors for CIOs managing large-scale IT operations.

The company's commitment to innovation ensures enterprise customers have access to cutting-edge semiconductor technology, enabling them to stay ahead of the curve in an increasingly data-driven business landscape. AMD's roadmap aligns well with emerging enterprise needs, including AI and machine learning capabilities, which are becoming integral to many business processes.

For IT directors concerned with vendor lock-in, AMD's x86 architecture provides a viable alternative in the CPU market, potentially increasing negotiating leverage and flexibility in procurement strategies. Additionally, AMD's recent acquisition of Xilinx expands its offerings in the FPGA space, providing solutions for specialised computing needs in sectors such as telecommunications and industrial automation.

AMD emphasizes robust ecosystem support, offering comprehensive resources for system integrators and IT teams. This includes detailed documentation, developer tools, and enterprise-grade support channels, facilitating smooth integration into existing IT infrastructures and maximising the potential of AMD's products in complex enterprise environments.

• Everything you need to know about Intel
• Everything you need to know about Nvidia

AMD has confirmed plans to acquire Finnish company Silo AI in a deal valued at $665 million. 

Based in Helsinki, Silo AI provides end-to-end AI solutions that enable enterprises to easily integrate the technology within products, services, and operations. The company works with clients spanning a host of industries, including Philips, Rolls-Royce, Unilever, and Allianz.

The firm’s open source large language models (LLMs), including Poro and Viking, are multilingual and created on AMD platforms alongside its internal SiloGen model platform.

As part of the deal, Silo AI chief executive and co-founder, Peter Sarlin, will continue to lead the company as part of AMD’s Artificial Intelligence Group. In his new role, Sarlin will report to AMD senior VP Vamsi Boppana.

“At Silo AI, our mission from the start has been to build an AI flagship company. Today's announcement is a logical next step in that pursuit as we join forces with AMD to shape the future of AI computing,” Sarlin said.

“We have a well-established history of building successful AI products and delivering value to our customers. We look forward to becoming part of AMD to further scale our impact and develop enterprise solutions and AI models that address the most complex challenges with deploying AI at scale today.”

AMD's acquisition spree is underpinning its AI strategy

AMD said the agreement represents “another significant step in the company’s strategy to provide customers with end-to-end AI solutions based on open standards”. 

Incorporating Silo AI’s team within its AI Intelligence Group will enable the firm to bolster the development of tailored AI models, platforms, and solutions, Boppana said.

“Across every industry, enterprises are looking for fast and effective ways to develop and deploy AI solutions for their unique business needs,” he commented.

“Silo AI’s team of trusted AI experts and proven experience developing leadership AI models and solutions, including state-of-the-art LLMs built on AMD platforms, will further accelerate our AI strategy and advance the build-out and rapid implementation of AI solutions for our global customers.”

The deal also marks the latest in a slew of acquisitions and major investments by AMD to support its global AI strategy. In the last 12 months, the company has invested over $125 million in a dozen AI companies to further accelerate its end-to-end solutions portfolio.

In August 2023, AMD announced plans to acquire the French AI firm and long-standing partner, Mipsology. Speaking at the time, Boppana said the deal would enable the company to improve inference capabilities and further expand its AI software development capabilities.

News of this acquisition was followed by another deal to purchase open source AI software firm, Nod.ai, in October last year. In a statement at the time, the company said the deal would enable it to accelerate the development and deployment of AI solutions.

This particular acquisition marked a significant step for AMD in its continuing rivalry with Nvidia.

Nvidia offers similar software solutions aimed at accelerating the development of AI models. These, however, remain largely closed-source and are designed specifically to work with the chip giant’s own GPUs.

By contrast, AMD said at the time the deal “strongly aligns” with its current strategy of building out an “open software ecosystem that lowers the barriers of entry” for enterprises.

AMD’s Frontier supercomputer has retained its position as the fastest supercomputer in the world for the third year in a row. 

The announcement was made at ISC High Performance 2024 and highlights the firm’s continued high-performance computing leadership (HPC), according to AMD.

Housed at the Oak Ridge National Lab, Frontier is powered by AMD EPYC CPUs and AMD Instinct GPUs, and recorded a Linpack score of 1.2 exaflops based on the latest Top500 list. 

The list now features three new systems, including the El Capitan Early Delivery System (EDS), RZAdams, and Tuolumne - all of which came in at 46, 47, and 48 respectively on the list. 

These systems mark the first supercomputers on the Top500 list powered by AMD Instinct MI300A APUs, launched in December 2023.

Brad McCredie, corporate vice president and general manager for data center and accelerated processing at AMD, said this marks a significant moment for AMD and showcases its leadership in the HPC field. 

AMD now powers 1567 supercomputers on the Top500 list, marking a 29% increase on last year. The firm also boasts 157 systems on the Green500 list, which ranks the most efficient supercomputers in the world. 

“The AMD Instinct MI300A APUs are setting the pace for innovation, delivering leadership performance and efficiency for critical workloads sitting at the convergence of HPC and AI,” he said.

“We are thrilled with the progress shown by LLNL today with their early submissions using AMD Instinct MI300A APUs. We look forward to our continued engagement with the teams at HPE and LLNL for further optimizations to drive increased performance as we help bring up El Capitan.”

AMD is driving ahead in the exascale era

AMD and HPE collaborated to build El Capitan, marking the continuation of a technology partnership that started while developing the Frontier system. 

The company said that the debut of the El Capitan EDS, RZAdams, and Tuolumne systems demonstrates that the El Capitan system is officially on track to become the second supercomputer to surpass the exaflop barrier. 

"RZAdams is a critical addition to LLNL's existing and highly capable unclassified computing ecosystem,” said Bronis R. de Supinski, chief technology officer for Livermore Computing, LLNL. 

Performance boosts created by the AMD Instinct MI300A APUs mean that this system will be a “strategically significant resource” and help support a wide range of scientific research initiatives, de Supinkski added. 

“RZAdams is supporting essential opportunities for code porting, optimization and software development for some of the applications that will eventually run on El Capitan. It is also allowing our scientists to push the boundaries of discovery with greater speed and efficiency than previous unclassified systems.”

Over the last year, AMD EPYC processors and AMD Instinct accelerators have been used to power a raft of new supercomputers, the company revealed. 

This includes support for Italian energy company, Eni, which announced its new HPC6 supercomputer, powered by AMD EPYC CPUs and AMD Instinct GPUs. 

Upon completion, HPC6 will be one of the world’s most powerful supercomputers designed specifically for industrial applications.

Lenovo and the University of Paderborn also recently announced plans to build a new supercomputer powered by AMD EPYC CPUs. The installation of the system is expected to start in the second half of 2024.

Additional support from AMD last year included the use of AMD Instinct MI300A accelerators to power the Hunter system, housed as the High-Performance Computing Center of the University of Stuttgart (HLRS). 

Development of the University of Krakow’s new supercomputer, Helios, will be powered by 4th gen AMD EPYC CPUs

AMD launches new Alveo V80 Accelerator for HPC 

AMD also announced the launch of the Alveo V80 Compute Accelerator Card, which aims to provide optimized hardware-accelerated systems designed for a wide range of compute-intensive workloads.

This new accelerator is designed to overcome performance bottlenecks for compute-intensive workloads with large data sets typically encountered in HPC, data analytics, and network security applications. 

The Alveo V80 accelerator card is powered by the AMD Versal HBM Series adaptive SoC, and delivers up to a two-time increase in both memory bandwidth and logic density. 

Similarly, it can provide up to four times the network bandwidth compared to the previous generation, delivering “massive hardware parallelism” and 820GB/s memory bandwidth.

The system allows for greater hardware flexibility, supporting real-time acceleration of diverse and custom workloads and data types used in genomic sequencing, sensor processing, blockchain, and next-generation firewall and network security. 

Supermicro confirmed it will include the Alveo V80 accelerator with AMD EPYC CPU-powered Supermicro servers.

“Supermicro’s strategic and long-standing collaboration with AMD enables us to address customer requirements for grappling with compute-intensive and memory-bound workloads at scale with superior time-to-delivery,” said Michael McNerney,  senior vice president for marketing and network security at Supermicro. 

“By leveraging the AMD Alveo V80 accelerator alongside our A+ server family, including the AS-4125GS-TNRT, customers can immediately take advantage of this compact 4U server solution for deployments where compute density and memory bandwidth are critical.”

AMD and Microsoft have expanded their partnership to bolster end-to-end compute and software capabilities for the hyperscaler’s enterprise customers. 

Unveiled at Microsoft Build 2024, AMD solutions such as its Instinct MI300X accelerators, ROCm open software, Ryzen AI processors and software, and Alveo MA35D media accelerators will support Microsoft offerings such as the Azure OpenAI service and its new virtual machines. 

The deal will help underpin a wide array of critical services from Microsoft, according to AMD president Victor Peng. 

“The AMD Instinct MI300X and ROCm software stack is powering the Azure OpenAI Chat GPT 3.5 and 4 services, which are some of the world’s most demanding AI workloads,” he said. 

“With the general availability of the new VMs from Azure, AI customers have broader access to MI300X to deliver high-performance and efficient solutions for AI applications.”

Kevin Scott, chief technology officer and executive vice president of AI at Microsoft, said the deal showcases the long-standing partnership between the two companies. 

“Microsoft and AMD have a rich history of partnering across multiple computing platforms: first the PC, then custom silicon for Xbox, HPC and now AI,” he said. 

“Over the more recent past, we’ve recognized the importance of coupling powerful compute hardware with the system and software optimization needed to deliver amazing AI performance and value. 

“Together with AMD, we’ve done so through our use of ROCm and MI300X, empowering Microsoft AI customers and developers to achieve excellent price-performance results for the most advanced and compute-intense frontier models. We’re committed to our collaboration with AMD to continue pushing AI progress forward.”

AMD supporting AI advances at Microsoft

First announced in preview in November 2023, the Azure ND MI300x v5 VM series is now available in the Canada Central region for customers to run AI workloads. 

Microsoft said these new VMs provide significant HBM capacity and memory bandwidth and will enable customers to fit larger models in GPU memory or use fewer GPUs. 

In the long term, this will help reduce power consumption and costs and rapidly accelerate time-to-solution. 

Powered by AMD’s ROCm software, the VMs are also being used for Azure AI production workloads. Microsoft said this will give customers access to GPT-3.5 and GPT-4 models. 

Hugging Face sees performance improvements with ND MI300X VMs

One of the first customers to harness the new VMs was Hugging Face, Microsoft and AMD revealed. 

The AI company ported their models to the ND MI300X VMs in just one month, which delivered significant performance and price/performance improvements for their models.  

As part of the deal, ND MI300X VM customers can now bring Hugging Face models to the new VMs to easily create and deploy NLP applications.

“The deep collaboration between Microsoft, AMD, and Hugging Face on the ROCm open software ecosystem will enable Hugging Face users to run hundreds of thousands of AI models available on the Hugging Face Hub on Azure with AMD Instinct GPUs without code changes, making it easier for Azure customers to build AI with open models and open source,” said Julien Simon, chief evangelist officer, Hugging Face.

Optimizing with AMD Ryzen AI software

AMD also announced that developers can now use Ryzen AI software to optimize and deploy inference on Ryzen-powered PCs. 

This software allows applications to run on the neural processing unit (NPU) built on AMD XDNA architecture, which is the first dedicated AI processing silicon on a Windows x86 processor.

This addresses traditional issues when running AI models on a CPU or GPU alone, which can drain battery power rapidly. However, with a Ruzen AI powered-laptop, AI models can operate on the dedicated NPU, which enables users to free up CPU and GPU resources for other tasks. 

AMD said this helps to both increase battery life while allowing developers to run on-device AI workloads and concurrent applications more efficiently. 

Advancing Video Services and Enterprise Compute

Finally, Microsoft revealed it selected the AMD Alveo™ MA35D media accelerator to power live streaming video workloads. This includes Microsoft Teams, SharePoint video, and others, the firm said. 

The Alveo MA35D media accelerator is purpose-built to power live interactive streaming services “at scale”, according to AMD. The company said this will enable Microsoft to ensure high-quality video experiences for users by streamlining video processing workloads, including video transcoding, decoding, and encoding. 

Using the Alveo MA35D accelerator in servers powered by 4th Gen AMD EPYC processors, Microsoft will unlock the ability to consolidate cloud Infrastructure by harnessing high channel density, energy-efficient, and ultra-low latency video processing capabilities. 

Similarly, the Alveo MA35D features ASIC-based video processing units that will support AV1 compression standards, enabling AI-enabled video quality improvements to provide a “smooth and seamless” video experience for users. 

AMD also noted that the deal will support ‘future-ready AV1 technology’. By supporting emerging standards like AV1, the Alveo MA35D “provides Microsoft with a solution that can adapt to evolving video processing requirements.” 

Semiconductor giant AMD has expanded its EPYC server CPU portfolio with the introduction of its new AMD EPYC 4004 Series of processors.

Aimed at powering entry-level system designs, the cost-optimized CPUs provide enterprise-class features and leadership performance suited to small and medium-sized businesses, as well as hosted IT service providers. 

Perks include enterprise-grade performance, dependability, and scalability alongside modern security features that are powered by AMD’s Zen 4 architecture.

In an announcement, John Morris, corporate vice president of AMD’s Enterprise and HPC business group, said the offerings will help deliver cost savings to organizations while also better catering to their hardware and performance needs.

“Historically, many small to medium businesses have had to compromise on their IT solutions by using hardware that doesn’t fully meet their needs,” he explained.

“Based on the same technologies that power the most demanding data centers in the world, the AMD EPYC 4004 Series processors are offered at an optimized acquisition cost for customers in small and medium-sized businesses seeking to drive better business outcomes.”

Under the hood of AMD's new EPYC 4004 Series

AMD’s new EPYC 4004 Series processors offer a single-socket package designed to tackle strong, general-purpose computing, enabling customers with cost and infrastructure restraint considerations to leverage highly-performant rack scale, multi-node, and tower configurations.

As an example, AMD said a server powered by a single EPYC 4564P CPU will deliver 1.8x increase in performance per CPU dollar when compared to an Intel Xeon E-2488 processor.

The EPYC 4004 Series includes eight variants, ranging in price from $149 for the EPYC 4124P, to $699 for the EPYC 4564P and the 4584P. 

AMD said the chips enable a range of broadly deployed enterprise solutions and are supported by the firm’s stable of partners – which include Altos, ASRock Rack, Gigabyte, Lenovo, MSI, New Egg, OVHcloud, Supermicro, and Tyan.

Commenting on the announcement, Kamran Amini, vice president and general manager for Server, Storage and Software Defined Solutions at AMD partner Lenovo, said the new CPU range will “help address new market opportunity.”

“The expansion of the AMD EPYC processor roadmap addresses a larger market segment with the affordable, yet high-performance capabilities of the new processor,” he said.

“With the introduction of AMD EPYC 4004 series CPUs, AMD is helping to address the needs of server customers across the spectrum, providing price per performance value for growing businesses along with efficiency for a dependable, secure, infrastructure.”

Optiver has partnered with AMD to make use of a range of high-performance compute engines to transform its data center infrastructure. 

The market trading firm will use AMD EPYC processors, AMD Solarflare Ethernet adapters, AMD Virtex FPGAs, and AMD Alveo adaptable accelerators as part of the deal.

The move will enable it to unlock marked performance enhancements and operational agility, according to Alexander Itkin, Optiver’s US CTO. 

“We’re a team of researchers and engineers driven by a common vision: building systems that make financial markets better,” he said. 

"We are constantly pushing the limits, whether it’s challenging the speed of light to minimize latency or building world-leading computational scale. We need computing infrastructure that enables us to navigate the complexities of the capital markets in nanoseconds. 

“For us, this is only possible with high-performance products from AMD, which help us successfully and continuously adapt to financial markets."

Archana Vemulapalli, corporate vice president for commercial enterprise at AMD, said the move will help Optiver to “push the limits of what is possible” with regard to data center performance and efficiency. 

“AMD is uniquely positioned to meet the needs of commercial customers with a robust and diverse portfolio of compute engines that enable impressive performance and energy efficiency; two critical components of the modern data center.”

How AMD products are supporting Optiver’s goals

As part of the deal, Optiver will use 4th Gen AMD EPYC processors to expand  large-scale compute required to drive data analysis and deliver core density for its data center modernization and workload consolidation program.

The market trading firm will also use AMD’s Solarflare X2 series Ethernet network adapters. These are specifically designed for high-performance electronic trading environments, and provide low-level control of network interfaces at the application level. 

Optiver said this will enable it to develop better-designed, scalable systems. 

Virtex UltraScale+ FPGAs will also enable Optiver to accelerate data capabilities and improve operational agility. 

“The exponential growth of data demands smart network and data center solutions that reach beyond fixed-function silicon devices to deliver maximum throughput, high data processing capability, and the flexibility to adapt to evolving connectivity standards,” AMD said. 

Kevin Sprague, US hardware lead at Optiver, said the use of Alveo Accelerators will be critical in enabling the company to “move more information, faster”. 

Optiver will utilize the AMD Alveo U55C and Alveo UL3524 adaptable accelerator cards as part of the deal. The latter of these is capable of delivering 7x latency improvements compared to previous FPGA technologies, according to AMD. 

“Our ultimate goal was to get faster access to data and information that is critical to meet the unique needs and workflows that the financial industry demands,” Sprague said. 

“Our collaboration with AMD helped us develop the products and features our industry demands. Additionally, our joint work has allowed us to expand and optimize our offering by leveraging not only AMD FPGAs, but also AMD EPYC CPUs to build out high-performance trading systems and infrastructure.”

AMD is celebrating major industry recognition after being awarded the 2024 Corporate Innovation Award by the Institute of Electrical and Electronics Engineers (IEEE). 

The semiconductor giant was recognized for its development and deployment of chiplet architecture designs aimed at transforming high-performance computing (HPC), which the institute described as pioneering. 

The industry recognition highlights the transformative research conducted by AMD’s engineering teams, according to Mark Papermaster, executive vice president and chief technical officer at AMD. 

“Several years ago, our passionate, world-class engineering teams began addressing the scaling challenges of traditional chip design,” he said. 

“Everyone takes modular architecture for granted now, but at the time it was radical thinking. Its success is as much a feat of engineering as a testament to the power of teams with diverse skills and expertise working together, despite uncertainty and risk, toward a common vision.”

AMD was among the first companies to harness the capabilities of chiplet architecture to drive advances in processor performance, efficiency, and flexibility, helping extend and build upon previous performance gains under Moore’s Law. 

The firm’s research has helped extend and build upon previous performance gains under Moore’s Law, and in 2018 introduced the first processor that featured a chiplet-based x86 CPU design.

The second-generation AMD EPYC server processor was powered by the “Zen 2” core architecture, which has since been built upon to deliver marked performance gains. 

AMD has also incorporated chiplet architectures across the breadth of its processor portfolio, including AMD Ryzen CPUs for PCs, AMD EPYC processors for data centers, and AMD Instinct accelerators for use in HPC and AI development. 

This marks the second time AMD has received this award. In 2005, the institute recognized the company for its pioneering research in the evolution of x86 microprocessors and its extension to a 64-bit architecture, which is the foundation of modern data centers and PCs.

How AMD led the way in chiplet architectures

AMD’s new processor design approach involved overhauling traditionally monolithic SoC designs into multiple smaller “chiplets”, which it said allowed researchers to rapidly accelerate time-to-market and deliver high performance, energy efficiency, and cost-effective chips. 

These chips leveraged common designs for multiple markets and a wide array of use cases. 

IEEE said AMD’s chiplet architecture has “nearly halved” the manufacturing cost of producing modern processors, and have brought notable advantages such as greater data bandwidth to memory, and higher compute capability compared to more conventional designs. 

AMD’s research into chiplet design continues to drive advances in computing performance and efficiency, which IEEE said will enable the industry to address some of the world’s most pressing challenges through scientific research. 

The chiplet-based AMD EPYC processors, for example, are currently used within the Frontier supercomputer, the world’s first exascale machine. Scientists are currently using Frontier to develop innovative new technologies for use in energy and medical science.

A GPU memory vulnerability dubbed ‘LeftoverLocals’ could expose LLM responses to hackers through leftover local memory, researchers have warned. 

Apple, Qualcomm, AMD, and Imagination are among the big name GPU vendors named as vulnerable according to research posted on the Trail of Bits blog. The GPUs have all been vulnerable to varying extents from as far back as September 2023, when researchers first began their investigation.

Researchers were able to build a proof of concept (PoC) of the potential attack, citing an attached video in which they listen in to another user’s interactive LLM session by recovering a GPUs local memory.

From their tests, the researchers concluded that ‘LeftoverLocals’ could leak around 5.5mb per GPU invocation on an AMD Radeon RX 7900 XT, enough data to reconstruct an LLM with worryingly high precision.

With GPUs used extensively to support the high performance requirements of AI inferencing, this news is likely to raise serious concerns among enterprise AI users.

Eleanor Watson, IEEE member and AI ethics engineer at Singularity university, told ITPro that although this particular vulnerability would require physical access to a GPU, the research highlights serious data protection risks.

“Whilst this particular type of exploit requires direct access to the GPU and its memory, it’s indicative of the challenges of keeping our interactions with AI systems private,” she said.

“I expect that further vulnerabilities will be uncovered which are broadly applicable to a wide range of LLM systems, leading to a widespread doxxing of interactions and generations, and associated embarrassment,” she added.

Vendors have acted swiftly on GPU memory vulnerability  

To defend against the vulnerability, GPUs need a built in system to clear local memory between kernel calls. Some GPU vendors, including Nvidia and Intel, are already doing this, whereas others need to keep pace. 

In response to the news of ‘LeftoverLocals’, AMD stated plans to create a new mode that “prevents processes from running in parallel on the GPU and clears local memory between processes on supported products”, with an expected rollout of said mode by March 2024.

“This mode would be designed to be set by an administrator and not enabled by default,” the firm said in an advisory. “Supporting documentation for the new mode, along with details of how to update AMD products, will be provided in a future update to this security notice.”

Apple responded to Trail of Bits but did not issue specific details of their patch, while Qualcomm issued a partial fix and Imagination a full patch in December.

A failure to patch this vulnerability could see attackers targeting a variety of GPU applications and LLM sessions, including those within privacy-sensitive domains.

Open source LLMs, however, are still the main target. Despite their ability to be rigorously audited, their reliance on closed-source GPUs means they are particularly susceptible to this attack.

“A lot of security hardening will need to be done on AI systems to make them more resistant to these kinds of vulnerabilities”, Watson said.

“This is a necessary component of the ongoing professionalization of AI, along with mastering challenges such as confabulation/hallucination in models”, she added.

When it comes to managing your IT estate, peak performance isn’t always about investing in the latest or the fastest technology. For many organizations, maintaining the infrastructure they already have in place and making sure it is what’s best for them is equally important. 

For most SMBs, tight budgets mean making shrewd investments in hardware and software. The aim is to buy the most suitable equipment for now, but not something that will be outdated when the business starts to grow. 

There is an element of risk here as decision makers are essentially predicting what they will need for the future and guessing how and when they might scale up. Pick a system that’s too small and they may find themselves spending again to upgrade. Alternatively, something large and expensive might not be worthwhile in the short term and could also be outdated before any return on investment is fully realized. The answer isn’t to aim for somewhere in the middle; no, it’s about what an organization can afford, and ultimately what you need. But most importantly, it’s about right-sizing. 

The art of right-sizing

“Right-sizing” is a practice commonly seen with cloud computing where resources are tailored to the precise needs of workloads. It takes away the ill-conceived notion that bigger is better or that the fastest products are the most efficient and simply leaves you to assess what is uniquely best for you and your business. The saying ‘one-size fits all’ should only really be for stretchy clothing and has no place in business IT as every setup has a unique data capacity and performance requirement, regardless of how flexible or agile the infrastructure is.

So-called forklift upgrades – the cumbersome process of moving data from one platform to another – should also be avoided. That becomes more challenging if there is more than one system or unit that needs replacing or upgrading. By selecting the right equipment for their operational needs and making sure it’s compatible with the other systems in their network IT decision makers almost guarantee their IT estate runs at its best and also reduce the chance of needing to make avoidable changes down the line. 

Patience is key; decision makers shouldn’t be rushed into a decision or fooled by systems that promise to deliver five years from now. They should focus on the next one-to-two years and assess where they are now before making choices that look too far ahead. Do they have remote workers? Are they sticking with on-prem servers? How much of the budget can they afford to spend right now? There will be many considerations to take into the equation, and they too may change over the next couple of years. If you go too big now, you might find it hurts other parts of your IT estate down the road. 

There’s also the problem of asymmetric lifespans. For example, a business could be kitted out with every piece of relevant hardware and software in one fell swoop on the same day, but the time each element needs to be refreshed or updated will vary by years or even decades

IT asset management (ITAM) is the administrator’s friend here. An IT asset is defined as hardware, software systems, or information (data) that an organization values. IT assets, regardless of their quality, price, or level of manufacturer support will eventually have an end of life. Within a network of other systems, that may come sooner than expected as different assets may need help to keep up with more modern, or powerful, products. 

Security should also be a high priority for SMBs; cyber threats have become increasingly complex and old, unpatched hardware is low-hanging fruit to the threat actors of today. When it comes to older hardware, the question shouldn’t just be “can it still do the job?” or even “can it still do the job effectively”, but “can it do the job safely?”. If not, for the sake of the entire business’ security the answer must be to replace it. 

Hyper-converged infrastructure (HCI)

The private cloud is often the answer; when it comes to servers, virtualization can help you increase capability without taking up more floor space. Hyper-converged infrastructure (HCI) is a software-defined IT infrastructure that virtualizes elements of hardware systems. It can replace siloed compute and storage equipment with standard servers and local storage, often with financial gains over the longer term. HCI can also be used for networking. VMware vSphere+ is an example of that as it creates virtual compute nodes, storage clusters, and switches that replicate a complete data center without any need for legacy architecture.  

For Hyper-converged infrastructure, AMD’s EPYC processors are a flexible option, with an array of core counts – from 8 to 64 – and frequencies that allow businesses to optimize their environments more efficiently. The 3rd Gen AMD EPYC processor is packed with up to 64 cores, allowing the user to consolidate many workloads onto fewer servers, which helps save energy, space, and money. 

AMD EPYC processors offer energy-efficient x86 servers and deliver exceptional performance and reduced energy costs. The result is maximum value for a limited IT budget. What’s more, it includes AMD Infinity Guard, which is a comprehensive suite of security features to keep your data safe. 

Maintaining your IT estate can be made considerably easier if you use right-sizing from day one, and a shrewd processor investment can give you the efficiency you and your budget need.

AMD has announced optimistic projections for the coming fiscal year, driven by commitments for its new AI chip platform, but its battle to overturn Nvidia’s AI market share is just beginning.

As the firm enters its final quarter of 2023, it has targeted strong growth in its data center chips range driven by increasing demand for the powerful hardware necessary to inference generative AI models. 

Key to this is AMD’s MI300 series of graphics processing units (GPUs), with the MI300X in particular offering enterprises the performance they require. 

Lisa Su, CEO at AMD, told investors that the company had projected $2 billion in revenue driven by MI300 sales in FY24, with $400 million in Q4 2023 alone. The news came amidst mixed results for the firm, with embedded and gaming segments seeing 5% and 8% against its client segment revenue’s 42% year-on-year rise. 

“We delivered strong revenue and earnings growth driven by demand for our Ryzen 7000 series PC processors and record server processor sales,” said AMD Chair and CEO Dr. Lisa Su. 

“Our data center business is on a significant growth trajectory based on the strength of our EPYC CPU portfolio and the ramp of Instinct MI300 accelerator shipments to support multiple deployments with hyperscale, enterprise and AI customers.” 

AMD’s data center division saw flat revenue across Q3, which the firm pinned on diminishing sales of system-on-chip (SoC) products. But it believes this division can be a major source of growth, with rapid uptake of enterprise generative AI driving fierce demand in the space. 

Without providing names, the firm stated that it is already ramping up shipments of MI300 chips in response to purchases from hyperscale and enterprise customers.

LLMs such as OpenAI’s GPT-4 or Meta’s Llama 2 contain billions of parameters, and training or inferencing models of this size is only achievable for most firms by paying to use bespoke hardware from providers like Nvidia and AMD. 

AMD targets AI battle with Nvidia

As in the gaming space, AMD’s main competitor in the AI hardware space will be Nvidia. The chips giant has had a bumper year, with its Hopper-based H100 chips becoming sought after for the training and inference of large language models (LLMs). 

 In terms of raw performance, AMD’s MI300X chip can be configured for up to 192GB of HBM3 and 5.2TB/s of memory bandwidth which the firm says is sufficiently powerful to handle an 80-billion parameter model on a single GPU instance. 

It compares favorably with Nvidia’s H100, each instance of which comes packed with 188GB HBM3 memory and a peak memory bandwidth of 3TB/s.

But Nvidia’s next-generation GH200 chip, built in anticipation of frontier AI models with a trillion parameters, is expected to be released in Q2 2024 and will present AMD with a tough standard to beat. 

Comparing individual chips only gives part of the picture when it comes to each company’s ability to handle massive AI workloads. 

Both firms have respective technologies that allow for individual GPUs to be linked in order to tackle the largest compute challenges. Nvidia’s NVLink connects H100s to form the 640GB memory DGX H100 and eventually GH200s to form the 144TB memory DGX GH200.

AMD’s Infinity Architecture Platform, meanwhile, can combine up to eight MI300X chips for a combined 1.5TB memory.

"The biggest stumbling block that any vendor will need to overcome to compete with NVIDIA in AI is the install base of NVIDIA’s CUDA development environment," James Sanders, principal analyst, cloud and infrastructure at CCS told ITPro.

"Multiple vendors have attempted to compete on cost, with startups like Graphcore struggling as the infrastructure savings for lower-cost / higher-performance AI acceleration does not outweigh the technical friction introduced by adopting these alternatives.

"Wider adoption of open-source libraries (e.g., PyTorch) will ease this for inference, though training is likely a sticker problem. Even with corporate-sponsored collaboration in displacing CUDA using open-source tooling, bringing alternatives to feature parity and training scientists in how to use alternatives a multi-year endeavor."

Where customers place their workloads will come down to a range of factors. However, it’s clear that when it comes to sheer performance AMD has reason for confidence, but it will have to offer competitive pricing if it wants to capture a slice of the market already using Nvidia’s technology.

The firm would benefit from a partnership with prominent firms, but could struggle as Nvidia has already partnered with Dell, as well as Snowflake, Microsoft, and Oracle. 

Other contenders in the space

AMD will not only have to contend for space in the AI market with Nvidia, but a number of other potential competitors.

AWS has leaned heavily into its own family of AI chips, Inferentia and Trainium, and is set for a collision of its own with Nvidia. With a focus on its internal capability to serve the compute needs of its clients, AMD will have to look at the other two of the ‘big three’, Microsoft and Google, if it hopes to score a decisive victory among the largest in the AI space. 

Customers of the Amazon Bedrock platform could see little reason to move outside of the AWS ecosystem, though the firm’s commitment to open AI development could give AMD space for third-party inference of models. 

Intel wants to achieve AI dominance by 2025, and has specifically aimed to corner the market when it comes to data center processing. 

Its Gaudi 2 GPUs were specifically designed with LLMs in mind, and the chip was specifically praised by Hugging Face machine learning (ML) engineer Régis Pierrard for its superior performance over Nvidia’s A100 GPU. 

Intel has also announced a wide range of hardware products aimed at capturing the AI market outside of CPUs, including its Granite Rapids and Sierra Rapids CPU family and field programmable gate arrays (FPGAs). 

Whether AMD can face off against more than one of its long-time competitors in order to find success in the AI space will depend on the versatility of its GPU lineup, and the ease and price of access for its services.

Running data centers is undeniably expensive, with costs including the hardware, software, energy consumption, and skilled labor to configure and maintain the infrastructure. While the value of running and maintaining on-premise IT infrastructure is clear, initial up-front and recurring costs are forcing many businesses to re-think their approach to infrastructure.

This includes snubbing surging cloud costs entirely, in favor of running applications and hosting data on-site. For example, the likes of US software firm 37 Signals, which embarked on a cloud repatriation journey in February.

To make matters worse, the costs of running data centers have fluctuated markedly over the past decade – falling, then rising sharply again – with concerns over energy prices and the need for greater sustainability dominating the conversation. In fact, data centers are among the most energy-intensive buildings on the planet, and consume up to 50 times the energy per floor space of the average commercial building, according to the US Office of Energy Efficiency and Renewable Energy.

Significant changes in the landscape, driven by rising energy costs, mean running data centers is becoming increasingly unsustainable for many organizations, particularly those that are unable to realistically pass all costs on to their customers. To ensure data centers still provide value, businesses must review their configurations so they are as efficient as possible, which includes the components that power them.

The impact of rising costs

Inflation is a difficult pill to swallow for data center operators, especially because it’s a reversal of a trend we’ve seen since the 2010s. Uptime research estimates an average Tier III enterprise data center would have cost $12 million per MW in 2010, with these costs as much as halving immediately prior to the COVID-19 pandemic.

Now, costs are once again sky-rocketing. The combination of rising construction costs and lead times, alongside a surge in staffing costs, is significantly affecting the viability of running data centers. In fact, two-thirds of enterprises and colocation operators are now anxious over higher spending on data centers in 2023, according to Uptime’s findings, which were published last year.

The sector not only faces price escalation but the uncertainty of supply, too, according to the 2022 Data Center Cost Index by Turner & Townsend. The demand for metals and energy has hiked the costs of construction materials and components, with longer lead times resulting from demand outpacing supply. The overwhelming majority (95%) questioned by Turner & Townsend reported delays to projects caused by shortages, with lead times for many major pieces of equipment extending by at least 12 weeks.

Energy prices in Europe have been particularly painful, with costs in the UK, for example, surging 635% between January 2021 and March 2022, according to FTI Consulting. The same has been seen in France (400%), the Netherlands (364%), and Germany (276%). Prices in Singapore, meanwhile, rose by 247%.

The majority of respondents (78%) to the 2022 Data Center Cost Index reported that securing reliable power supplies has become a significant blocker to data center development. Indeed, as is the case in Frankfurt, concerns over data center sprawl have led to government action to curb growth and ensure a more even distribution of resources.

“Governments, investors and large data centre operators alike expect sustained action to drive greater operational efficiency – reducing pressures on stretched power grids and limiting environmental impact,” the report adds. “The future may see green finance being the only finance option coming forward, requiring demonstrable, clear sustainability plans to attract new investment.”

Powering a sustainable future

While the regulatory picture may gradually result in the enforcement of pro-environment policies, data center operators must also take charge of the issues blighting the industry in a bid for energy efficiency.

While public sector policies and the rise of renewable energies might inject some hope into the macro picture, data center operators can also take a number of steps to tackle the issue. Reviewing the components that power one’s infrastructure, and assessing the market for more energy-efficient processors, such as the AMD EPYC series of data center chips, are both an ideal place to start.

AMD EPYC CPUs offer excellent performance, and power the most energy-efficient x86 servers, with EPYC-fitted infrastructure minimizing the environmental impact of data centers while contributing to the organization’s sustainability goals. Testing of 4th Gen AMD EPYC processors – the latest series in this family of CPUs – shows that including these chips in your data centers leads to 67% fewer servers required and a 52% power reduction, versus competitive CPUs.

The result is a reduction in capital expenditure of up to 40% in the first year, alongside an operational expenditure reduction of 61%.

With data centers continuing to expand globally, and rising energy prices fuelling greater expenditure, organizations should do everything they can to shrink costs. By investing in technology like energy-efficient AMD EPYC processors, businesses will be able to reduce their power consumption and pivot towards net zero, without sacrificing performance or their compute capabilities.

To find out more information about the potential of 4th Gen AMD EPYC processors, visit AMD's official website.

Microsoft is reportedly working on its own chips for training AI models, which could help expand its cloud AI offerings and tighten competition with rivals AWS and Google Cloud.

The move could allow Microsoft to further expand its generative AI products, such as those launched through its partnership with OpenAI, and embed AI tools across its productivity ecosystem.

A Bloomberg report stated that the chip is codenamed ‘Athena’, and could be the product of a strategic partnership between Microsoft and AMD.

Microsoft’s need to train large language models (LLMs) at scale has sharply increased over the past year, as the company placed generative AI processes front and center in its future.

AMD shares rose after the initial reports, though neither firm has officially announced a partnership of this kind.

Microsoft spokesman Frank Shaw told Bloomberg that while “AMD is a great partner,” it is not involved in Athena.

“Microsoft’s reported move into custom silicon makes sense, particularly as Microsoft is integrating AI inference into its own products, like Windows, Bing, and Office,” said James Sanders, principal analyst, cloud and infrastructure at CCS Insight, to ITPro.

“There’s also ample precedent for this, as Amazon Web Services designs and offers the Trainium and Inferentia chips, and Google Cloud builds and offers TPUs, while Google at large also builds custom silicon for video encoding to support YouTube.”

Google has claimed that its AI chips are ‘faster and greener’ than competitors, with the firm having stated that its 4th generation tensor processing unit (TPU) can train AI at 1.3-1.9 times lower power intensity than Nvidia A100 chips.

Sanders explained that although Copilot and Bing AI will generate revenue through subscriptions and advertisements, they will also drive workload requirements to the extent that Microsoft could save money by using in-house hardware.

“While Copilot features are paid add-ons and Bing with AI is ultimately ad-supported, Microsoft will have the volume of inference workloads to warrant building custom chips for it, and doing so will improve the margins associated with adding AI functions to these products. 

“Depending on how far Microsoft is in development, it may also help to hasten the scale of availability to more users, as Nvidia is experiencing sustained demand for A100 and H100 chips.”

He noted that Microsoft is highly likely to continue to provide Nvidia graphics processing units (GPUs) for its cloud customers, in the same manner that Amazon has provided Intel and AMD CPUs alongside its Graviton Arm CPUs.

In November, Microsoft announced a collaboration with Nvidia to build an AI supercomputer, a years-long project involving tens of thousands of Nvidia’s AI chips.

Where does big tech stand on AI hardware?

Big tech is flocking to Nvidia at present, with Microsoft, AWS, and Adobe all having chosen the firm’s cloud infrastructure for LLM training.

Microsoft has been a particularly large customer, having worked with Nvidia on the aforementioned supercomputer in addition to having paid Nvidia significant sums as part of its $10 billion OpenAI investment.

OpenAI has relied heavily on Nvidia’s cloud architecture and hardware to train its GPT-3 and GPT-4 LLMs.

It has particularly benefited from its CUDA software framework, which has helped to provide a consistent development platform that firms can rely on to be compatible with their software.

AMD could benefit hugely from focused Microsoft investment and assistance, which could place it in a much better position to shake Nvidia from its top spot.

Nasdaq figures for 2022 showed that Nvidia controlled 88% of the GPU market compared to AMD’s 8%.

Despite this, AMD demonstrated stronger growth in 2022, with revenues rising 44% year on year, driven in large part by data centers, compared to Nvidia’s 0.2%.

Intel has also targeted AI hardware dominance by 2025, with a roadmap for its diverse product range seeking to address current bottlenecks such as memory bandwidth.

The key to the chip giant’s success could lie in its open approach to AI, having helped Hugging Face improve the open-science BLOOM LLM using its Gaudi 2 processor.

In a blog post, Hugging Face engineer Régis Pierrard compared Intel’s hardware favorably to Nvidia’s A100 AI chip and stated that the former offered better performance and lower costs.

Intel’s existing customer base may also help it quickly scale its AI market share, with approximately 6.2 million active developers in its community ready to adopt its AI tools and hardware.

Using its tool SYCLomatic, developers can migrate CUDA source code to Intel’s open-source SYCL programming model which could ease the way for a steady stream of customers moving away from Nvidia.

ITPro has approached Microsoft and AMD for comment.

Apple CEO Tim Cook has revealed the company will be among the first customers at TSMC’s new US-based fabrication facilities.

The Taiwan-based company announced plans yesterday to triple its investment in US operations from an initial $12bn to $40bn and build a second production facility in Arizona.

The move marks one of the largest foreign investments in US history and has been hailed as a major milestone in bolstering domestic chip-making capabilities.

TSMC chairman Mark Liu revealed that the first fabrication plant is on course to be operational by 2024 while the second plant is expected to be up and running by 2026.

Liu added that the fabrication plants will deliver around 13,000 high-quality jobs and are expected to produce 600,000 chips each year - enough to meet growing demand from American manufacturers.

Once operational, the fabs will produce some of the most advanced chips currently available, including the four-nanometer and three-nanometer chips which are used to power Apple’s M-Series and A-Series processors, along with a host of Nvidia and AMD products.

“Exciting Day for America”

Tim Cook attended the ceremony alongside executives from Nvidia and AMD. In a speech at the event, he described the investment as an “incredibly exciting day for America, for Arizona, and for Apple too”.

Cook revealed that Apple will be among the first companies to purchase chips manufactured at the Arizona facilities. In a tweet afterwards, he added that the tech giant will be the plant's “largest customer”.

“We work with TSMC to manufacture the chips that power our products all over the world, and we look forward to expanding this work in years to come as TSMC forms new and deeper roots in America,” he said.

“Today is only the beginning. Today we are combining TSMC’s expertise with the unrivalled ingenuity of American workers. We are investing in a stronger, brighter future. We are planting a seed in the Arizona desert, and at Apple we are proud to nurture its growth,” Cook added.

AMD chief executive Lisa Su echoed Cook’s comments, confirming that the company expects to be a “big customer of both fabs”.

“We could not do what we do without our partnership with TSMC,” Su said. “Which is why we are so excited about today’s milestone."

American manufacturing capabilities

President Biden welcomed the TSMC investment as a significant moment for US chip manufacturing. The US once produced around one-third of the world’s computer chips. However, in recent years this has dipped to around 10%.

This has forced companies such as Apple, Intel, and Nvidia to rely heavily on overseas manufacturing to meet growing demand for products.

“Today, we’re down to producing only around 10 percent of the world’s chips, despite leading the world in research and design in new chip technologies,” Biden said in a speech at the Arizona facility.

“The United States is better positioned than any other nation to lead the world economy in years ahead if we keep our focus,” he added.

The Biden administration and its predecessor both signalled intentions to accelerate domestic production amid global supply chain disruption and rising tensions with China.

In August, the Biden administration signed into law the CHIPS and Science Act. The act aims to lure manufacturers to US soil and will allocate more than $52bn in grants, loans and incentives to support manufacturers such as TSMC.

Both TSMC factories in Arizona will be partially subsidised by the US government, the Biden administration confirmed.

Despite positive remarks from Biden and key industry stakeholders, Mark Lippett, CEO for AI chip designer XMOS, told IT Pro that reducing reliance on overseas manufacturing and ensuring self-sufficiency in the US represents a “real challenge”.

“The semiconductor industry is so vast and so complex that it doesn’t really work without a globalised world,” he said. “If there was a global directive that mandates no chip sales to China at all, for example, the majority of the semiconductor businesses would go belly-up.”

“There’s also the complexity of the supply chain to contend with. It’s a concatenation of hugely expensive, complex concepts, from very specific IPs to bespoke and specialised equipment. You can’t just throw money at the problem with specific expectations for ROI,” Lippett added.

Creating a more sustainable business has become a key priority for organisations – for instance, the majority (60%) of Fortune 500 companies have climate and/or energy efficiency goals. They’re looking at their operations, their supply chains, their offices and how they make and deliver their products or services and looking to see how they can make them more cost and energy efficient. Inevitably, the enterprise data centre has a huge part to play in this drive.

The intergovernmental organisation, the International Energy Agency, believes that data centre electricity use now makes up over 1% of global energy demand. By 2025, they could produce 3.2% of global carbon emissions. Adopting green data centres – where construction and operations minimise energy and water consumption and use alternative energy sources and cooling methods where possible – can help drive this environmental impact down.

Public awareness of the environmental impact of data centres is growing, and governments are also taking notice. The EU is expanding or reviewing legislation and initiatives, including the Ecodesign Regulation on servers and data storage products, the EU Code of Conduct on Data Centre Energy Efficiency, and the EU Green Public Procurement criteria. Its Proposal for a Directive on energy efficiency brings in new elements designed to improve the energy efficiency of data centres. In the UK, the UK Carbon Reduction Initiative sets out strict rules for data centre managers if their facility uses more than 6,000MW per year of power, while the Department for Environment, Food & Rural Affairs (DEFRA) has described data centres as a ‘key area of focus.’

While data centres pose real challenges and opportunities for companies trying to operate sustainably, it is possible to make them greener.

Designing greener

How? Well, the ideal approach is to start from scratch, building sustainability into every aspect of the design and construction. This could even start with where you locate the data centre, selecting a position that’s already cool, not inclined towards humidity and more likely to benefit from a good, natural airflow.

Of course, this isn’t always possible, but every data centre can be designed with air and cooling management in mind. Cooling data centres can take up to 40% of their total energy consumption, and power usage efficiency (PUE) metrics already take this into account. By using natural airflows in cooling and avoiding the recirculation of warm air, data centre operators can reduce energy use and associated emissions.

Dividing data centres into hot and cold aisles can make a difference, managing intakes and exhausts to ensure that hot air can be vented from the hot aisle and cool air introduced through the cool aisle, for the most efficient configuration. Doors and walls can also direct airflow, while isolating the equipment that produces the most heat can help, particularly with measures to get that heat out of the building as rapidly as possible. New air-conditioning systems may be more energy efficient than existing ones, not just reducing their environmental impact, but their long-term operating costs.

What’s more, immersive cooling systems, like those designed and manufactured by Asperitas, transfer heat to a synthetic oil through natural convection, reducing the energy waste and costs of cooling and enabling the heat to be used elsewhere. They also make it possible to reduce the physical footprint of the servers and infrastructure, allowing data centres to be located where they would previously have been impractical.

Greener power

There are also opportunities to cut waste and consumption in how the power is supplied. Before reaching servers, network infrastructure and storage racks, the power typically moves through multiple uninterruptible power supplies (UPS) and power distribution units (PDUs). What’s more, many data centres use multiple power sources and UPS systems in order to provide a level of redundancy. While these generally run at a high level of efficiency, this can vary; a modern modular UPS or PDU might achieve 96% to 99% efficiency under a full load, but older or less capable units might be closer to 94%, wasting significant amounts of energy over time. Units with energy efficient modes and intelligent management are more likely to push this up into the 98% to 99% range.

In some cases, shifting to an energy supplier that uses more renewable energy sources, or even connecting your own renewable sources, could also reduce greenhouse gas emissions (scope 2). Unlike Amazon, not every business can operate a wind or solar farm to power their data centre, but there may be a way to harness renewable resources in a more modest and accessible way.

Crucially, organisations need to make the most of tools and technology that allow them to monitor and measure power consumption, then optimise based on the data. Intelligent PDUs, data centre smart assistants and smart power and temperature controls all have their part to play. Obviously, Google works at the largest scales, yet it’s pioneered the smart approach with smart temperature controls and data-driven insights from its DeepMind AI tech. By combining the two, Google was able to reduce its data centre energy cooling power consumption by 40% over 18 months.

Going deeper

All these measures will help to build a greener data centre, but to maximise their energy efficiency, organisations also need to go deeper at the server hardware level. You can start by looking at how servers and racks are cooled, using variable speed fans that react to temperature or even liquid cooling for devices running more demanding, high-performance applications. You can integrate energy efficiency into your storage strategy, using SSDs, with their reduced heat output and power consumption, where you need performance, and slower, more efficient hard drives where you don’t.

Most importantly, you need to use the most energy-efficient processors and server hardware and optimise how and when you use them as much as possible. Again, using data and AI-based management tools can help by dropping servers into sleep states outside peak hours of demand, but the key here is really virtualisation. Consolidating workloads from physical servers to virtual servers reduces power consumption for the servers themselves alongside the power consumption of the systems that cool them. When the multinational bank, DBS, moved from virtualising 50% of workloads to 99% in its private cloud, it shrank the physical footprint of its servers in one data centre to a quarter of its original size, and reduced its power consumption by half.

Part of this success came down to a shift to AMD’s EPYC server CPUs. With more cores and a higher RAM capacity per server, DBS was able to run more virtual servers from a single physical machine. AMD’s own research has demonstrated that ten AMD EPYC 7713 can reduce energy consumption by 32% and save an estimated 70 metric tons of greenhouse gas emissions. What’s more, with more cores per socket and more cores per server, it’s easier to fit more processing power into a nimble dual or single-socket server, reducing server counts and footprints even further.

AMD’s current generation EPYC processors already produce exceptional levels of performance per Watt. Yet AMD has even bigger ambitions for the future, committing to a 30x increase in energy efficiency for AMD processors and Instinct GPU accelerators by 2025, representing a 2.5x acceleration of the industry trends from 2015-2020, and a 97% reduction in energy use per computation from 2020-2025. You only have to look at The Oak Ridge National Laboratory’s Frontier supercomputer to see what might be possible. Each of its 9.408 HPE Cray EX nodes combines an AMD 7A53 EPYC CPU with four AMD Instint MI250X GPUs and 9.2 petabytes of memory. It’s the world’s fastest supercomputer, with a theoretical peak performance of 1.686 exaflops, yet also the world’s most energy-efficient supercomputer, with 52.23 gigaflops of performance per watt. That’s 32% more efficient than the previous energy-efficiency leader.

And note the focus on accelerators. As more organisations ramp up investments in big data and AI applications, energy-efficient GPUs will be as vital as optimised CPUs in reducing power consumption and tackling climate change. If all AI and HPC server nodes matched AMD’s promised power reductions, billions of kilowatt-hours of electricity could be saved every year. As we move away from an approach to computing focused on the CPU to one that utilises a wider range of accelerators, adaptive compute engines, data processing units and FPGAs, we’ll see these more application or workload-specific processors deliver even more performance per watt.

Building a more sustainable data centre is a holistic enterprise, taking in everything from the physical space to how individual servers are configured, cooled and managed. But by building on the latest processor technology and optimising how it's used through virtualisation, organisations can put themselves in the best position to meet their sustainability goals. Businesses need to look beyond cores, threads and raw performance, and focus more on efficiency when making hardware choices, but with AMD’s technology, the architecture means you can have both – and reduce TCO.

If you’re looking for more detail on how, tools like AMD’s AMD EPYC bare metal server TCO estimation tool can help

It’s no secret that server hardware updates can be onerous, expensive and disruptive. When your small business has a small IT team and limited resources, it’s tempting to hold off for as long as possible and do your best to maintain the hardware you’ve got. Yet often the smarter choice is to grasp the nettle, find the budget and push ahead. A hardware update can help you reduce long-term costs and risks, but also bring new opportunities to rethink what your servers and supporting infrastructure can do.

In terms of costs and risks, the arguments are simple. Firstly, as servers age, they become more expensive to maintain and run. IDC studies have shown that older servers with lower performance and higher failure rates can increase unplanned downtime by 20%, while server performance degrades by an average 14% per annum, meaning that, after five years of life, it’s 40% slower than when it came out of the box. Older servers aren’t just more expensive to maintain, but also less energy efficient. They may run warmer, needing additional cooling in the server room. This also drives up total cost of ownership (TCO).

Old, inefficient servers can eat up IT resources which could be better used elsewhere; a problem for small businesses, where limited budgets and manpower are the norm. They can also cause issues for performance, stability and availability that impact productivity elsewhere. What’s more, as they reach (or go beyond) their warranty period and end of life, further risks emerge. Software or firmware may no longer be patched to cover vulnerabilities or meet emerging threats. There may be compatibility or performance issues with new applications that your business wants to run.

Take a piecemeal approach, replacing hardware only when it goes beyond repair, and you end up managing a more complex, mixed estate with different hardware and different management tools. Again, this can be daunting for small businesses working to manage that estate without access to a wide range of specialist expertise.

Updating your servers with new hardware can help you minimise these inefficiencies and risks. What’s more interesting, though, is that it could also create new opportunities and help you control your operating costs.

For a start, new server CPUs offer dramatic performance and efficiency gains. AMD’s third generation ‘Milan’ EPYC processors brought AMD’s powerhouse Zen 3 cores to the server marketplace, not to mention improvements in cache, memory bandwidth and IO throughput. These processors offer an up to 2.2x performance increase over the second generation EPYC CPUs, and the performance delta over previous generation Intel Xeon processors is even larger; your existing servers may have CPUs from several generations ago. The range spans everything from single-socket systems with eight-core processors to dual socket systems and 64-core CPUs. Whether you’re a small business with modest but growing requirements or a high-tech startup hungry for processing power, EPYC CPUs have the performance, features and scalability you need.

Of course, it’s not just the CPU that drives performance gains, but also improvements to the platform as a whole. If your existing servers rely solely on magnetic hard drives for storage, new hardware that uses M.2 SSDs for primary storage will be significantly faster. This goes double with the PCIe 4.0 SSDs supported by AMD’s architecture. You may also have access to higher RAM speeds of up to 3200MHz DDR4, and even higher quantities of RAM than your current hardware can support.

It’s also worth mentioning that AMD doesn’t limit its strongest feature set to its two socket server platforms. If you’re operating on a budget, you can take on a single-socket server and still gain all the benefits of the third generation EPYC architecture. Beyond the obvious drop in purchase costs, this could potentially save SMBs a fortune in software licences. When budgets are tight, even smaller savings count.

The point of the exercise isn’t just to run existing workloads faster, but to allow you to do more within a small estate. New platforms with more cores per CPU could host more containers and VMs, giving you the ability to serve more users with more applications, scale upwards when you need to - or even consolidate the workloads you already run. With more cores per CPU, more cache and more RAM, you could virtualise more servers and do more work on fewer physical boxes, using less energy for power and cooling, in less space.

In tests, AMD has demonstrated ten AMD EPYC 7713 servers handling the same workload as 15 servers running the Intel Xeon 8380, consuming an estimated 32% less energy and with an estimated 42% lower TCO. The Xeon 8380 is a recent, third-generation Intel chip. Replacing servers based on older hardware, it’s not unheard of to see organisations reducing their existing server count by up to 75%. That’s a huge plus for sustainability, not to mention TCO.

Virtualisation doesn’t just equal consolidation. With more performance and more headroom for virtualisation, you’ve got the scope to scale up applications on demand, or pilot new applications without splashing out on additional hardware, as and when you need to. There may be new applications in data analytics, machine learning or AI that could bring value to the business, but that your existing servers won’t run.

Plus, by embracing hyperconverged infrastructure, businesses gain the flexibility to manage computer and storage resources across multiple servers in a cluster, making it easier to scale out workloads and use automation. With HCI, even smaller IT teams can run a powerful infrastructure with the capability to do much more than their existing servers ever could. Indeed, vendors can deliver certified HCI configurations designed and built specifically for this job. With new tools and management features, IT teams can manage their estates from a single dashboard, and free up resources for more business-focused tasks.

Updated hardware is also more secure. Not only can you look forward to years of patches and firmware updates, but also new security features. The growth of CPU exploits such as Spectre, Meltdown and Foreshadow has shown CPU security to be critical, and some older architectures to be vulnerable. Newer architectures are designed to resist these attacks, while incorporating additional protections to meet new threats.

All AMD EPYC CPU models feature AMD Infinity Guard technology to secure data in system memory DIMMs, CPU registers and cache. Secure Encrypted Virtualisation (SEV) technology, which encrypts each virtual machine running on the CPU with a unique encryption key, is currently exclusive to AMD. This encryption happens on the fly and with minimal-to-no performance degradation. Supported by many popular hypervisors, it allows your existing applications to run unmodified inside secure VMs. Meanwhile, to ensure system integrity, the server firmware is also protected by AMD Secure-Root-of-Trust technology, to prevent malicious code from booting at startup. You’re not going to find this level of security on older server CPUs.

Perhaps most crucially, updating your hardware can give you a new competitive advantage – the chance to embrace a modern, agile server infrastructure that can power new applications. These applications can help the business find new customers, deliver stronger, more personalised produces and services and reduce product development lifecycles and time to market. They can help them do more with smaller teams and limited resources. Older architectures were designed for the old IT. Modern architectures are ready for what’s next. In short, don’t look at hardware updates as an unfortunate but necessary exercise, but as an opportunity to rethink your IT and make sure your business has access to the tools it needs to thrive and grow.

For a clear idea of how new hardware could help you manage your estate and reduce your TCO, take a look at AMD’s EPYC server virtualisation TCO estimation tool

Perhaps the only thing more confusing about buying a new business laptop - or an array of them for an entire workforce - besides the sheer amount of choice on the market, is the myriad different components that equip each model. Laptops vary wildly in terms of technical specifications, so it’s important to understand how each will impact your workers’ ability to get their jobs done.

The central processing unit (CPU), sometimes simply referred to as the ‘processor’, can be seen as the heart, eyes, ears, and brain of the laptop, and is arguably the most important component of the entire device. The confusion here, again, lies in the amount of choice on offer. There are so many different models from different manufacturers on the market, it can feel overwhelming to choose one, especially when the machine will have a direct impact on the quality of work your business is able to produce.

It also doesn’t help that manufacturers don’t make it easy to truly understand what each CPU model will provide you in terms of performance. Many often stick with old hardware conventions, opting for initialisms and numbers in the thousands rather than something a little more explicit. Users often find themselves trawling through comparison websites and social media recommendations before ultimately landing on the CPU they end up purchasing.

That said, there are a few things to know before buying a laptop and understanding what makes a good processor could lead to a better, more productive purchase that stands the test of time, no matter what your business can throw at it.

The best laptop processor

Put simply, there is no straight answer to the question ‘what is the best laptop processor?’ The answer depends on a multitude of factors, many of which relate to the type of work for which you plan to use the business laptop. It will also depend largely on your budget, as laptop CPUs can get eye-wateringly expensive the more ‘premium’ you go.

Before buying a laptop, you need to have a strong idea of what you need the machine to do. Running software for general office work like Microsoft Office 365 is the base standard level of functionality most businesses expect, but some will demand more depending on the job role. For example, you’re unlikely to want to equip an office administrator with the same device as a senior software engineer because the computational demands of a programmer will most likely far exceed those of a more general office worker.

As an easy rule, the best laptop CPU for your business is almost always going to be the most powerful chip you can afford. So, set a budget, understand what you need from the device, and take things from there. But before you go shopping, knowing the differences between manufacturers is hugely important as it could affect performance and compatibility in the long term.

Is Intel or AMD better?

There are two main manufacturers of laptop CPUs on the market today and they are AMD and Intel. Like the allegiances held among devout Xbox and PlayStation console owners, the fanbase for each company can be equally loyal at times, and which is better is ultimately a matter of preference and timing.

Timing is key when making a business laptop purchasing decision, especially when it comes to CPUs. This is because the two companies have been locked in battle for years, each being considered the best manufacturer at different times. Historically, and now again in the present, Intel is considered to make the faster CPUs thanks to its latest 12th-generation Alder Lake chip architecture. However, AMD stole Intel’s crown for a few years recently with its Ryzen 5000-series line of CPUs. These were considered to be vastly superior chips to Intel’s 10th and 11th-generation chips at the time, so if you’re planning to buy an older laptop, or a used machine, then it would be best to check the dates and model numbers to ensure you’re not getting a sub-optimal setup.

CPU naming conventions can be confusing but, fortunately, when it comes to filtering out the Intel chips to avoid, the company’s model numbers clearly specify which generation of chips the model comes from. Each model number starts with the number corresponding to the generation of chips it's from, so an ‘Intel Core i7-12650HX’ processor tells you it’s from the latest 12th-generation line of chips because the model number begins with 12.

A quick way to differentiate the two manufacturers in today’s market is to say that Intel offers better performance, but AMD’s CPUs are more power-efficient and won’t run as hot - generally speaking. Regardless of which manufacturer you choose, you’ll get a highly capable device thanks to modern technology, but if you want the very best on the market, Intel takes the top spot… for now, anyway.

How to choose between an Intel or AMD processor

Intel

Intel’s consumer-grade CPUs come in four main tiers:

• Not the least powerful chip Intel produces but the least powerful we recommend for business use. This is functional for basic office software but not anything more computationally demanding, like photo editing software, for example.
• The happy middle ground that will suit most businesses. Laptops equipped with an Intel Core i5 will breeze through basic software and run more demanding programs well in some cases, but others may experience some lag or stuttering.
• The i7 range is a favourite among the enthusiast crowd and is powerful enough to handle almost any software a business could need with ease. These models are typically much more expensive than an i3 or i5, so they should only be considered if your budget is large or if there is a genuine business need for such performance.
• A more recent addition to the consumer-grade lineup, the i9 is unlikely to be found on many business-focused laptops, but is one of the best models available if your business needs laptops to run many computationally demanding tasks all at the same time, for example.

AMD

Handily, AMD’s naming structure for its flagship line of chips is very similar to Intel’s, in that the most affordable series is indicated by the number 3 in the Ryzen 3 line, and the best performance is offered by the Ryzen 9 chips. It’s worth noting that the most recent Ryzen 6000 chips eliminate the 3-series, however, offering just Ryzen 5, 7, and 9.

Each tier can broadly be characterised in the same way that Intel’s can, too, which is to say the Ryzen 5 is like the i5 - they are both the happy middle ground, while the Ryzen 7 and i7 are both high-end chips that will run almost anything well. The full list of AMD’s consumer-grade CPUs that we’d also recommend for business are as follows:

• AMD Ryzen 3
• AMD Ryzen 5
• AMD Ryzen 7
• AMD Ryzen 9

If you’ve already done some research into AMD Ryzen chips, you may have encountered the company’s AMD Ryzen Mobile Processors for Business line of chips. These may appear to be different to the normal Ryzen Mobile CPUs, but the chips themselves are the same.

Intel vPro vs AMD Ryzen Pro

However, there are also AMD’s Ryzen Pro Mobile Processors for Business, and Intel has its own equivalent line called vPro. These chips are almost identical in terms of performance, but are aimed at businesses who want additional features that could improve things like cyber security and management, for example.

Both Intel vPro and Ryzen Pro use the same model naming schemes as described previously, but offer features like additional security layers to prevent cyber security vulnerabilities. Both AMD and Intel have worked with Microsoft to offer Secured-Core PCs which protect against firmware-level cyber attacks and offer extra provisions for securing sensitive data - somewhat important given today’s data regulatory climate.

The extra features that come with these lines of chips could be hugely beneficial to your business so it’s worth having a conversation with your IT team to decide if they are worth the extra money, and the time it would take to configure for an entire workforce.

Business laptop CPUs to avoid

Intel and AMD also have other consumer-grade CPUs on the market but no others we would feel comfortable recommending for business use. Intel has its Pentium and Celeron lines which, in our opinion, do not offer enough performance to run a business machine seamlessly, and all the software it typically requires. Equally, AMD’s Athlon CPUs will offer performance that will likely leave your workers needing more to get the job done well.

The potential savings to be made by choosing a lower-spec model over the ones we recommend can seem attractive, but when you consider how long you want the machine for and factor in degrading performance over time, it’s better to invest more at the outset rather than having to replace the laptop a year down the road.

Laptops are now essential to modern business. They offer flexibility, mobility and high end functionality which enables employees to meet their organisation's needs and deliver on key objectives; in fact, it's all but impossible to imagine what a modern business would look like without access to notebook PCs.

Their significance has become even more acute since the coronavirus pandemic led to an enormous surge in remote working. Without laptops, many businesses would have been unable to continue operating during the extended lockdown periods, and even now that restrictions have lifted, remote working is firmly embedded as an accepted norm.

This surge in remote working has fed into a growing desire from businesses of all stripes to pursue digital transformation and innovation in order to establish a competitive edge, but the responsibility for enabling these technical capabilities is often placed in the hands of IT teams, with little extra resources (if any) to support this increased scope. This added burden has put more pressure than ever on the time and workloads of technicians.

Streamlining and time-saving

One of the biggest challenges for any IT team is simply keeping the wheels turning and the lights on, ensuring that all an organisation's infrastructure and equipment is functioning normally. However, while the regular monitoring and maintenance of IT systems is essential work for keeping a business’s technology operational, the more time that can be freed up from performing these tasks – which are often repeated, manual processes – is time that can be spent on enacting some of the more ambitious transformation initiatives that may deliver increased value overall.

This is why enterprise device management solutions such as the tools built into the AMD PRO platform have become one of the cornerstones of business IT. They allow organisations to administrate all of the laptops in their fleet from a single, centralised dashboard, quickly identify problems and begin working on remediation efforts. This helps speed up workflows and increase efficiency, allowing smaller teams to manage larger numbers of employee devices. With technical talent in such high demand, this means that the growth of your business won’t be held back by the growth of your technical support staff.

When a new Windows update is released by Microsoft, for example, IT admins can automatically roll it out to all devices in the organisation on a pre-set schedule to minimise disruption, while ensuring a speedy upgrade without having to install it machine by machine. Rolling out upgrades manually on an individual basis is a long, tedious process, and the ability to take care of the update process systematically using native tooling frees up large amounts of IT time while granting access to the latest features and security improvements. Similarly, if for any reason the update needs to be rolled back, this can also be quickly and easily accomplished without having to physically recall devices.

This means that issues can be resolved much faster, which is particularly important for fast-paced modern businesses. After all, downtime can be expensive, and in most organisations, an employee without access to their laptop is an employee that has little means of being productive. Enterprise manageability tools help IT teams monitor employee devices so they can proactively fix a range of potential hardware and software problems before they turn into major issues. They include the ability to remotely change settings, apply software fixes and even trigger system restores to roll the device back to a previous stable recovery point.

Secure in the knowledge

Asset inventory is another key capability. Knowing exactly what devices are in your fleet – as well as their hardware configuration and firmware status – is a vital part of maintaining a robust security strategy, and the ability to invisibly gather this data and centrally log it without having to manually audit devices is one of the principle reasons why enterprise manageability tools are so important. In addition, they also help assist with compliance practices, and allow for more efficient coordination of warranties, support contracts and device refresh plans.

Because AMD PRO’s management tools are designed to integrate seamlessly with Microsoft Endpoint Manager, organisations can leverage these abilities without having to adapt their existing device management stack to accommodate them. what's more, because they're based on open standards and DASH profiles, there's no risk of vendor lock-in, and a comprehensive range of third-party enterprise mobility management solutions are also supported to ensure that IT teams can manage their AMD PRO-equipped fleets as easily as possible alongside the rest of their devices with a consistent experience.

All of these capabilities are available remotely too, which is essential for organisations that need to support workers both at home and in the office. Hybrid working is now a fact of life for the majority of businesses, and having the right tools to address problems from anywhere ensures not just that staff can continue working without having to grapple with technical hiccups, but also ensures they feel happy and looked after, which translates to better employee satisfaction and reduced turnover.

Alongside the robust platform validation, long-term software stability and multi-layered hardware-based security protections, these features have been designed to make AMD PRO a processor platform built for the needs of modern, agile business computing.

Read AMD’s ‘The challenge: Doing more with less’ to learn more about enterprise manageability

Just as there’s a direct link between the performance of a business and the performance of its employees, so there’s a link between employee productivity and the hardware they use on the job. Slow application performance, unresponsive software or an unreliable PC can all have consequences. A 2020 UK study found that nearly half of British workers felt that outdated tech hindered their productivity, and that slow technology cost them, on average, up to 46 minutes a day. With a speedy, responsive PC, however, users have time to focus and work creatively, rather than wonder how long they’re going to have to wait while their PC grinds its gears.

This makes it crucial that IT teams refresh their PC and laptop fleets regularly. Yet this raises another issue: How do you evaluate performance and make meaningful comparisons that can guide you towards the right choice?

Evaluating PC performance has rarely been simple or straightforward, particularly when it comes to CPUs. Back in the 1990s you could make some assumptions based on clock speeds; the more MHz you had to play with, the faster your applications ran. The problem is that while many businesses still make these assumptions, maybe factoring in core counts, threads and – possibly – cache, these methods of assessment are fundamentally flawed.

As a new white paper from AMD, ‘Measuring what matters – Benchmark considerations for commercial PC purchases’, makes clear, modern processor design is less focused on ramping up clock speeds, and more focused on creating efficient super-scaler designs with deep instruction pipelines and a more sophisticated use of cache, enabling each core to execute multiple instructions per clock cycle. In fact, designers may even focus on reducing the number of clock cycles required to execute a given set of instructions, rather than rely on pushing clock speeds upwards. If you automatically assume that a processor running at 3 to 4.8GHz is going to give you more performance than one running at 1.9 to 4.4GHz, you could be making a big mistake.

In fact, AMD’s white paper suggests, “The idea of boost and base frequencies is an oversimplification – the actual frequency at any given time will depend on workload, power-mode and platform thermal design.” This is true of desktop PCs, but even more so of laptops, where the thermal design of the device will constrain how long the CPU will be able to run at its maximum boost frequency. In real world usage, one CPU in one laptop might reach a higher boost frequency, but quickly throttle to a lower speed as power demands increase and temperatures rise. Another CPU in another laptop might reach a lower boost frequency but prove able to sustain it over longer periods of time.

Meanwhile, core counts matter. Provided an application is designed to make effective use of multi-threading – and most are these days – the AMD CPU with eight cores running 16 threads will still run a given operation faster than the Intel CPU running four cores with eight threads, higher clock speeds notwithstanding. In fact, it might find it easier to stay within thermal constraints and sustain higher clock speeds longer. Throw in all the other variables within the system – memory bandwidth, storage bandwidth, GPU performance – and simple MHz comparisons can be misleading.

So, how can businesses get a more accurate idea of real-world PC performance? The ideal would be a programme of user testing, with tests designed to mimic everyday real-world tasks in real working environments. The problem is that this is long, difficult to manage and potentially expensive, while it's difficult to pull out consistent and unbiased metrics. Software benchmarks can be more practical, but they need to be approached in the right way.

The obvious answer is application benchmarks, either using industry-standard PC benchmarks or, where the development resources are available, custom scripts based around real-world business tasks. However, both approaches have their issues. With industry-standard benchmarks, businesses need to make sure that they’re a good fit for the organisation, and not focused on consumer workloads or creative workloads that don’t reflect the kind of applications they actually use. Custom benchmarks, meanwhile, need regular updating to ensure that they’re not optimised for older architectures and still provide accurate, relevant results.

Both types share the problem that, while they can provide an accurate picture of performance in existing applications, they can’t predict how PCs will perform with new or emerging applications. What happens when a new, revolutionary tool appears that uses specific hardware features, but your benchmarks can’t tell you if a PC measures up?

This is where synthetic benchmarks come in. They can provide a snapshot of system or component-level performance, giving you some idea of how system X compares to system Y, or the read/write speeds of one SSD when compared to another. They can be quick, dirty and effective, but they need to be used with caution. As AMD’s white paper notes, narrow benchmarks of, say, the storage or memory subsystems may not call on features in the CPU pipeline or cache, skewing the result. “It is important not to overly weight these results,” it concludes, suggesting that “decision makers should instead evaluate the composite score, which exercises a broader set of processor functions”.

What’s more, synthetic benchmarks may not run for long enough or put enough stress on every part of the system to assess how performance will hold up against issues with memory bandwidth, I/O latency or thermal constraints.

It’s clear that no single benchmark or type of benchmark works for evaluating PC performance on its own, but where does that leave us? AMD’s white paper takes a smart approach, which combines “a wide range of both application-based and synthetic tests”. Through this kind of holistic, comprehensive testing, AMD shows that the Ryzen 7 Pro 5850U processor exceeds its rivals in a range of real-world applications, even though it runs at lower clock speeds and falls behind in some synthetic, single-threaded tests. No single benchmark can tell the whole story – and MHz to MHz comparisons won’t even get you past ‘once upon a time’. If you’re serious about evaluating PC performance, you really need the wider picture.

Read AMD’s ‘Measuring what matters – Benchmark considerations for commercial PC purchases’

AMD is set to acquire the networking vendor Pensando for around $1.9 billion.

Pensando offers a distributed services platform that includes a fully programmable packet processor and comprehensive software stack that it says accelerates networking, security, storage, and other services for cloud, enterprise, and edge applications.

AMD and Pensando said that its packet processor can be distributed throughout a network to accelerate multiple infrastructure services simultaneously, offloading workloads from the CPU and increasing overall system performance. In real world-cloud deployments, the companies underlined that Pensando’s product demonstrates between 8x and 13x greater performance compared to competitive products.

“We have seen a 40x improvement in overall cloud-based connection related performance. Pensando delivered this in less than 12 months,” said Girish Bablani, corporate vice president at Microsoft Azure, one of Pensando’s customers. “This partnership has enabled a transformational “limitless networking” shift that will ensure our customers have access to the highest performing products and services they expect from Microsoft Azure.”

Pensando was founded in 2017 and counts on several other high profile customers including IBM Cloud, Oracle Cloud, and Goldman Sachs.

“To build a leading-edge data centre with the best performance, security, flexibility and lowest total cost of ownership requires a wide range of compute engines,” said Lisa Su, AMD chair and CEO. “All major cloud and OEM customers have adopted EPYC processors to power their data centre offerings. Today, with our acquisition of Pensando, we add a leading distributed services platform to our high-performance CPU, GPU, FPGA and adaptive SoC portfolio.”

As part of the deal, Pensando CEO Prem Jain and the rest of the team will join AMD as part of the Data Centre Solutions Group. It is set to remain focused on executing its product and technology roadmaps, but now with additional scale to accelerate its business and address growing market opportunities across a broader number of customers.

“In less than five years Pensando has assembled a best-in-class engineering team that are experts in building systems together with a rich, deep ecosystem of partners and customers who have currently deployed over 100,000 Pensando platforms into production,” said Jain. “Joining together with AMD will help accelerate growth in our core business and enable us to pursue a much larger customer base across more markets.”

Following the customary closing conditions of the deal, the acquisition is expected to close in the second quarter of 2022.

Small form factor PCs make sense for businesses thanks to the combination of good performance and tiny enclosures, and the MSI Pro DP20Z 5M is a versatile new contender.

MSI’s latest machine is powered by impressive AMD Ryzen silicon, and it has plenty of ports and some good features. There’s plenty of wiggle room on the pricing front, too: it’s sold as a barebones system with two different processors, but it’s also available as a fully-built system in three configurations, so you’re able to use your own components or pick a specification that matches your workload.

The mid-range barebones model we’ve reviewed costs £299 exc VAT, which means it undercuts rivals. It’s also available as a £432 exc VAT full system with 8GB of single-channel memory, a 256GB SSD and Windows 10 Home.

MSI Pro DP20Z 5M review: Design

The MSI’s concentration on versatility continues to its enclosure – a boon if you’ve bought the barebones version of this device and need to fit your own hardware. The base and roof panels are secured with a couple of screws, so they’re easy to remove, and the internal layout is straightforward.

Beneath the roof you’ll find a basic processor cooler, two SO-DIMM memory slots and the motherboard’s first M.2 SSD connector, so users won’t have to detach the CPU cooler to install memory and storage. There’s a second M.2 port underneath the base, and a cage beside the motherboard holds pairs of 2.5in hard disks or SSDs. The MSI’s build quality is mediocre – it should withstand office life, but its metal panels and plastic façade could be stronger.

MSI’s rig weighs 1.52kg, so it’s reasonably light, and it’s 193mm wide, 161mm deep and just 85mm tall – so it’s not large when compared to conventional office desktops. That makes it easy to slot this machine into any tiny office space. It’s also compatible with 75mm VESA mounts, and adapters to convert to the more common 100mm format are easy to find.

Bear in mind, though, that competitors are even more compact. The Asus PN51 has two memory slots and sole M.2 and SATA ports and it’s smaller and lighter than the MSI. The Intel NUC 11 Pro crams three M.2 slots and two memory sockets inside an even more compact case. Even MSI’s own Pro DP21 11M is lighter and shorter.

MSI Pro DP20Z 5M review: Hardware and performance

Inside the MSI’s accessible chassis you’ll find an AMD Ryzen 5 5600G. It’s an impressive desktop APU, with six multi-threaded cores that use AMD’s excellent Zen 3 architecture. The chip has single- and multi-core peaks of 4.4GHz and 4.1GHz, and it uses an integrated Radeon graphics chip with 448 processing cores.

Our £299 exc VAT barebones machine has been tested with 16GB of dual-channel 3200MHz memory and a WD SN530 1TB SSD with mid-range read and write speeds of 2,441MB/sec and 1,996MB/sec – typical of the hardware that businesses will install in this machine.

The AMD chip helped the MSI deliver a superb benchmark result of 243. That’s miles ahead of the low-power laptop chips that Intel uses in NUCs. It’s also far better than the Intel Core i5-11400 inside MSI’s own DP21 11M, which only scored 158 in our benchmarks, and it crept ahead of the Ryzen 7 5700U used inside the Asus PN51.

The MSI’s integrated graphics core helped the rig deliver a 3D Mark Fire Strike score of 3,234. That’s decent and will help this machine tackle mainstream graphical workloads, and it’s better than Intel’s integrated GPUs. It handles low-end gameplay, too, which is ideal if you want to tackle some esports titles after work.

These benchmarks bode well for a wide variety of work situations. This machine will tackle any Office application, mainstream photo-editing apps and even some 1080p video-editing, and it won’t have any issue with multi-tasking. Indeed, it’s far better than any Intel rival in multi-threaded and creative tasks. The AMD chip never had thermal issues, either: the fan noise was barely noticeable even during tough tasks, and the processor didn’t show any throttling or temperature problems.

The only way you’ll get better performance with the DP20Z is by investing in the AMD Ryzen 7 5700G model, which delivers two extra cores and marginally better speeds. It’s not available as a barebones rig, though, so you’ll have to buy the full system for £582 exc VAT – but it’s hamstrung by 8GB of single-channel memory, and that will hinder speeds.

That £582 exc VAT could also buy you an entry-level Mac Mini, though. Its Apple M1 chip offers similar speeds to the 5700G with a smaller, more stylish case – that’s a tempting option, especially if you like macOS.

We still think the Ryzen 5 5600G barebones model is the sweet spot for this range. It’s faster and cheaper than the rival barebones system than Asus and the Intel NUC and it’s far quicker than MSI’s own Pro DP21 11M.

The DP20Z isn’t expensive to kit out, either: 16GB of dual-channel memory, a mid-range 256GB SSD and a copy of Windows will cost less than £200 exc VAT. That setup is a better option than the pre-built MSI: its £432 exc VAT price might undercut the barebones model once you’ve bought components, but its single-channel memory is slow. And, if you’ve got parts lying around, that’s an even better route to a cheap build.

The Ryzen 3 5300G barebones is good for low-end work tasks, too: it only costs £232 exc VAT, while the equivalent Asus costs £291 exc VAT.

MSI Pro DP20Z 5M review: Ports and features

The Ryzen 5 5600G is undoubtedly impressive, but the MSI’s features are underwhelming. The machine’s front edge has a USB 3.2 Gen 1 port, a USB 3.2 Gen 2 Type-C connector and two audio jacks. At the rear there are pairs of USB 3.2 Gen 2 and USB 2.0 sockets, and HDMI, DisplayPort and D-SUB outputs for triple-screen setups.

That’s reasonable everyday connectivity, and on the inside you’ll find dual-band Wi-Fi 6, Bluetooth 5.2 and Gigabit Ethernet. The Ryzen 3 versions of this machine, meanwhile, only have 802.11ac dual-band wireless. Security is handled by TPM 2.0, and the MSI supports 64GB of DDR4 memory at 3200MHz.

Those are fine features for everyday business computing, but there’s no denying that the MSI is missing some capability too. MSI’s machine doesn’t have super-fast USB ability and it has no Kensington lock slot, and the absence of PCI-E 4 means you can’t use the faster SSDs in this PC.

The Asus matches the MSI for memory and storage ability and it has better USB-C connectivity, a card reader, and an infrared sensor. It can also drive four displays simultaneously. The Intel NUC has Thunderbolt 4 ports for added speed, and both rivals have 2.5Gbps Ethernet. The MSI Pro DP21 11M has eight USB connections, but no Type-C connectivity.

MSI Pro DP20Z 5M review: Verdict

The MSI Pro DP20Z 5M isn’t the best when it comes to connectivity – it’s got the physical and wireless features to handle everyday workloads, but rivals offer faster networking and more ports. Competitors are smaller and lighter, too.

Look beyond this, though, and there’s plenty to like. Its AMD APU is one of the fastest chips you’ll find in any tiny PC, and in most of its barebones and fully-specified configurations it offers better value than the competition.

The lower price does mean you lose features and compact design, but the MSI Pro DP20Z 5M offers impressive power and versatility and it’s still very small. If you want compact speed on a budget, this is a good choice.

MSI Pro DP20Z 5M Specifications

AMD has set an April release date for its flagship V-Cache chip, along with several low-end processors to round out its Ryzen family.

The AMD Ryzen 7 5800X3D will ship on April 20 for $499, the company said. Unveiled in January, is the first Ryzen chip to use the company's V-Cache technology. It's an eight-core Zen 3 chip that stacks level 3 cache memory on top of the logic die, tripling it from the 32Mb on the Ryzen 7 5800X to 96Mb.

V-Cache is a technology that stacks 64Mb of SRAM level three cache on top of the computing die, connecting with it using a technology that AMD calls Through-Silicon Via. The company designed it to be transparent to developers, who can treat it like a regular chip with extra cache.

AMD executives have said that the 5800X3D will outperform both the Ryzen 5900X and Intel's Core i9-12900K chip. The company has targeted the product's performance increase at gamers rather than content creators, and claims 15% more performance than processors without the stacked cache technology.

The company also announced a set of six processors across its Ryzen 3, 5, and 7 families spanning its Zen 2 and 3 architectures. The units range from the four-core Ryzen 3 4100 with a 6Mb cache and PCIe 3 support at $99, through to the Ryzen 7 5700X, with eight cores and 36Mb of cache memory featuring Gen 4 PCIe support for $299. These chips will be available on April 4, the company said.

The AMD Ryzen 5800X3D provides a welcome high-end bump to the Zen 3 line before the Zen 4 range launches later this year. This will require a new motherboard featuring upgraded socket technology.

AMD also announced extended support for its Ryzen 5000 series processors running on AMD 300 series chipsets. This gives motherboards using the AMD X370, B350, and A320 chipsets support for these chips, including the desktop processors that it announced this week.