What is the AMD Enterprise AI Suite?

TL;DR

• AMD Enterprise AI Suite is an open source software stack designed to accelerate the deployment of production-grade AI models on AMD Instinct™ GPUs
• AMD Enterprise AI Suite is a modular suite that helps customers establish vendor independence
• AMD Enterprise AI Suite works through four components: the AMD Solution Blueprint, AMD Inference Microservices, AMD AI Workbench, and the AMD Resource Manager.
• AMD AI Workbench is an easy-to-use, low-code setup that can be deployed as a standalone product or combined with the AI Resource Manager

Businesses want to run LLMs and AI programs fast and efficiently, but are often trapped in closed ecosystems tied to just one vendor, cutting them off from the global AI community.

What they need is a platform built on open source foundations, with the resources and compute that truly transform their operations. Enter the AMD Enterprise AI Suite. This is a fully open source software stack designed to accelerate the deployment of production-grade AI models on AMD Instinct GPUs.

How open is the AMD Enterprise AI suite?

The AMD Enterprise AI Suite is an open, modular suite that enables you to maintain vendor choice, allowing customers to benefit from the global AI community. The suite is built on an established set of open source frameworks that allow for broad compatibility. Components of the stack can be deployed independently, using containers for flexibility, helping avoid dependency on just one vendor.

The suite is designed for seamless integration with existing DevOps and MLOps pipelines via Kubernetes and also supports on-prem and cloud deployments. The underlying acceleration is powered by the AMD ROCm software stack.

How quick is AMD Enterprise AI Suite?

AMD Enterprise AI Suite works through four components, each designed to accelerate deployment and, ultimately, maximize ROI.

It’s also versatile: not all of the components need to be used, with the modular design of the suite allowing customers to adopt only the ones they need, whether that’s Solution Blueprints, Inference Microservices, AI Workbench, AMD Resource Manager, or any combination of the four.

Solution Blueprints are prebuilt designs for enterprise use cases. Inference Microservices are optimized containers that offer fast deployments on AMD hardware. They feature support for open models and OpenAI-compatible APIs, along with intelligent hardware detection and tuning for maximum performance and cost efficiency.

AMD AI Workbench is an acceleration of AI development and collaboration, with prebuilt pipelines for fine-tuning foundation models and inference deployment, streamlining the journey from prototype to production.

How can AMD Enterprise AI Suite be used for LLM development?

AI Workbench is the key component of AMD Enterprise Suite for LLM development. The interface is specifically for developers to manage their AI stacks through an easy-to-use, low-code setup. IT can be deployed in different configurations, such as standalone – where features operate independently with basic authentication – or combined with AI Resource Manager.

AI Workbench in standalone mode ties to one project namespace, which is created during installation. All resources (AIMs, workloads, workspaces, and datasets) are then managed within the namespace.

However, when combined with Resource Manager, AI Workbench can support multiple project namespaces. Users can access projects they belong to, with AI Resource Manager handling the namespace lifecycle, as well as its quota allocation and organizational hierarchy.

AI Workbench comes with a range of capabilities, including an AIM Catalog, training and fine-tuning, GPU as a service, API Keys for programmatic access, and running workloads through the command line.

What resources are available in the AMD Enterprise AI Suite?

AMD Resource Manager optimizes GPU use and manages cost control with intelligent workload scheduling, quota management, distribution, and actionable telemetry – all of which enhance throughput, fairness, and capacity planning across teams.

AMD Resource Manager has an array of tools to oversee and control the platform’s computational resources and user access. The key capabilities include cluster management, monitoring, and maintaining teams access to computational resources.

Resource Manager is built around a basic usage pattern for maintaining compute resources, setting up projects, and allowing individuals to use the resources for their compute needs.

The suite is feature-rich, with clusters – the physical part of the platform installation – that can be managed in Resource Manager. Clusters include:

• Organization: this pertains to teams and can be used for multiple groups and projects.
• Quotas: these are a usage limit reserved for a project. These are useful for ensuring everyone gets a fair share of compute resources.
• Secrets: these are secure information, such as API keys or credentials that can be created at the organizational level and assigned to projects. These ensure workloads can access what they need without exposing your sensitive data.
• Storage: this can be configured to provide the project with the required credentials and connection information for workloads to access storage options like S3. Like secrets, storage configurations can be created at the organizational level and assigned to projects.