Optimize Large-Scale Training and Inference on AMD Instinct™ GPUs

Test drive AMD ROCm™ for open-source GPU programming with PyTorch and other leading frameworks. Find sample notebooks, tutorials, open-source projects, documentation, and more.

Install ROCm
See Supported GPUs

Run Sample AI Workloads

Try out the ROCm software stack for cloud and data center AI use cases. 

Access All AI Tutorials

Featured Workloads

Get Access to AMD GPUs

Want to Test Drive ROCm on AMD Instinct GPUs Instantly? 

Join the AMD AI Developer Program to get $100 in free credits for the AMD Developer Cloud — a powerful computing platform with AMD Instinct GPUs, pre-installed Docker containers, and zero setup. 

Sign up for the AMD AI Developer Program
Access the AMD Developer Cloud
AMD Instinct MI350 Series

Learning Resources

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

Community and Support

AMD AI Developer Program

Join the AMD AI Developer Program

Access cloud credits, connect with AMD engineers, receive early invitations to events, gain recognition opportunities, enter member only sweepstakes, publish technical content, and unlock exclusive training and workshops

Learn More

Other Forums

Pytorch related questions and issues
vLLM related questions and issues
ROCm related questions
Ready to file bugs or PRs?

Frequently Asked Questions

ROCm supports over a dozen Linux distributions for compute workloads on select AMD Instinct and AMD Radeon GPUs. Currently supported Linux distributions for ROCm are listed in the system requirements document.

Successive ROCm releases have seen significant leaps in performance. See the performance results page to get a reference point for evaluating how AMD Instinct GPUs perform with ROCm software running popular AI models for inference (e.g., vLLM, xDiT) and training (e.g., PyTorch, Megatron-LM, and JAX MaxText).

AMD releases official ROCm PyTorch Docker images regularly alongside new ROCm releases. These images undergo full AMD testing. See an overview, supported modules, and key libraries in the PyTorch compatibility document, and find instructions and downloads for the Docker image or a wheels package in the PyTorch on ROCm installation guide.

You can optimize your system by tuning AMD GPUs for your workload to leverage their full capabilities.

You can also use ROCm tools to improve the performance of AI models and speed up the inference stage of running AI workloads. Quantizing and pruning pre-trained models, optimizing computation kernels, and using accelerated libraries are methods of optimizing inference. Get started in the ROCm inference optimization document.

Join the AI Developer Program

Access free AMD Developer Cloud credits, exclusive training, monthly AMD GPU sweepstakes, and community support designed to support your AI development work.

Join Program