AMD Enterprise AI Reference Stack

Overview

Accelerate Enterprise AI adoption from prototype to production – built on open-source foundations and optimized for AMD Data Center Technologies.

By connecting key open-source AI frameworks and Gen AI models with an enterprise-ready Kubernetes platform, the AMD Enterprise AI Reference Stack of software components minimize the time from AI experimentation to large-scale production on AMD compute platforms.

Built for enterprise scale on AMD Data Center Technologies, intuitive interfaces enable organizations to quickly start with AI development and efficiently utilize compute resources across the organization.

AMD Enterprise AI Reference Stack at a Glance

• Accelerate AI adoption with immediate access to AI inference and training frameworks, developer tools and state-of-the-art AI models all optimized for AMD Instinct™ GPUs.
• Reduce costs with intelligent resource management and dynamic resource allocation that maximizes AMD compute utilization across the organization.
• Open and modular design: a fully open-source platform built on established open frameworks to avoid vendor lock-in.
• Enterprise-grade reliability with validated solution blueprints, optimized inference services, and an integrated developer workbench.

AMD Inference Microservices (AIMs) - Reliable, Optimized Serving

Prebuilt inference containers that bundle model, engine, and tuned configurations for AMD hardware. Supports OpenAI-compatible APIs and open-weights models, enabling rapid deployment without re-engineering. Hardware and use-case-specific tuning optimizes the deployment for best performance.

AMD Solution Blueprints — Reduce Risk, Accelerate Outcomes

Pre-validated reference implementations for common enterprise AI use cases shorten project timelines and reduce engineering effort. Blueprints provide repeatable, production-ready patterns for real-world workloads.

AMD AI Workbench — Empowering Application Developers and Data Scientists

The AI Workbench offers an intuitive, low-code interface to onboard and interact with AIMs. Users can deploy AIMs for production and integration in third-party applications without the need for deep AI or software developer expertise.

Workspace environments (JupyterLab, VS Code, ComfyUI) and a catalog of reference workloads for model training, fine-tuning and evaluation give data scientists and application developers the toolkits they need to quickly start building AI models and applications on AMD GPUs.

AMD Resource Manager — Control Costs and Utilization

Intelligent workload scheduling, guaranteed quotas, dynamic allocation of unused resources, and telemetry prevent GPU waste and ensure fair allocation. Single Sign On (SSO), isolated project namespaces, and Role-Based Access Control (RBAC) access to data and credentials enable adhering to enterprise security requirements across teams.

Get Started

Learning Resources

Explore more topics to find relevant learning resources to help you stay updated with the latest technological advancements from AMD.

Related AMD Technologies

Explore additional AMD platforms and developer resources that power the AMD Enterprise AI Reference Stack:

• AMD ROCm™ Software Platform - Open software stack for GPU compute and AI development.
• AMD Instinct™ Accelerators – High-performance GPUs optimized for AI training and inference.
• ZenDNN – Deep neural network acceleration library optimized for AMD EPYC™ CPUs.
• AMD Developer Central – Central hub for tools, SDKs, and performance guides. 

Disclaimers

The AMD Enterprise AI Reference Stack empowers organizations to accelerate AI adoption with open-source flexibility, optimized performance, and enterprise-grade control – enabling AI to move from experiment to impact.