AMD ROCm documentation

2024-12-20

4 min read time

Applies to Linux and Windows

ROCm is an open-source software platform optimized to extract HPC and AI workload performance from AMD Instinct accelerators and AMD Radeon GPUs while maintaining compatibility with industry software frameworks. For more information, see What is ROCm?

ROCm supports multiple programming languages and programming interfaces such as HIP (Heterogeneous-Compute Interface for Portability), OpenCL, and OpenMP, as explained in the Programming guide.

If you’re using AMD Radeon™ PRO or Radeon GPUs in a workstation setting with a display connected, review Radeon-specific ROCm documentation.

ROCm documentation is organized into the following categories:

Install

• ROCm on Linux

• HIP SDK on Windows

• ROCm on Radeon GPUs

• Deep learning frameworks

• Build from source

How to

• Use ROCm for AI

• Use ROCm for HPC

• Fine-tune LLMs and inference optimization

• System optimization

• AMD Instinct MI300X performance validation and tuning

• GPU cluster networking

• System debugging

• Use MPI

• Use advanced compiler features

• Set the number of CUs

• Troubleshoot BAR access limitation

• ROCm examples

Conceptual

• GPU architecture overview

• Input-Output Memory Management Unit (IOMMU)

• File structure (Linux FHS)

• GPU isolation techniques

• Using CMake

• PCIe atomics in ROCm

• Inception v3 with PyTorch

• Oversubscription of hardware resources

Reference

• ROCm libraries

• ROCm tools, compilers, and runtimes

• Accelerator and GPU hardware specifications

• Precision support

• Graph safe support