README

[!WARNING]
 This repository will be deprecated soon. Use
 ROCm/TheRock moving forward, including
 for issues and
 discussions.


  
  
    Open-source software stack for AMD GPU computing
  
  
    
      ROCm Core SDK
    
     • 
    
      AI Ecosystem
    
     • 
    
      GPU Systems and Infrastructure
    
     • 
    
      Blogs
    
  


 AMD ROCm™

ROCm is an open-source software stack of drivers, development tools, and APIs
that enable GPU programming from low-level kernels to end-user applications. It
is particularly well-suited to GPU-accelerated high-performance computing (HPC),
AI, scientific computing, and computer-aided design (CAD).

ROCm is powered by HIP, a
C++ runtime API and kernel language that lets developers write portable GPU code
with an interface similar to NVIDIA CUDA™. It also supports OpenMP and OpenCL,
and includes the compilers, debuggers, and libraries needed to build and run GPU
workloads. ROCm integrates with machine learning frameworks such as PyTorch and
TensorFlow.

 Table of contents

- Supported hardware and operating systems
- Quick start
  - Get started with ROCm
  - Deep learning frameworks on ROCm
- Core components
  - Math and compute libraries
  - Communication libraries
  - Runtimes and compilers
  - Profiling and debugging tools
  - Control and monitoring tools
  - Media libraries
  - Storage
- ROCm Extras
- Release notes
- Licenses
- Contribute

---

 Supported hardware and operating systems

Use the Compatibility matrix for official support across ROCm versions, operating system kernels, and GPU architectures (CDNA/Instinct™, RDNA/Radeon™, and Radeon Pro). Recent releases cover Ubuntu, RHEL, SLES, Oracle Linux, Debian, Rocky Linux, and more. GPU targets include CDNA4, CDNA3, CDNA2, RDNA4, and RDNA3.

---

 Quick start

Follow these instructions to start using ROCm.

 Get started with ROCm

Follow the ROCm installation guide to install ROCm on your system.

 Deep learning frameworks on ROCm

See Install PyTorch for ROCm or
Install JAX for ROCm to get started.

To learn more about AI training and inference workloads on ROCm, see the AI
Ecosystem
documentation portal.

---

 Core components

The core ROCm software stack consists of the following components. Most of them
are divided across the ROCm Libraries
and ROCm Systems super-repos by domain.

 Math and compute libraries

- Composable Kernel
- hipBLAS and rocBLAS
- hipBLASLt
- hipCUB
- hipFFT and rocFFT
- hipRAND and rocRAND
- hipSOLVER and rocSOLVER
- hipSPARSE and rocSPARSE
- MIOpen
- rocPRIM
- rocThrust
- rocWMMA

 Communication libraries

- RCCL
- rocSHMEM

 Runtimes and compilers

- HIP
- HIPIFY
- LLVM
- ROCr Runtime
- SPIRV-LLVM-Translator

 Profiling and debugging tools

- ROCm Compute Profiler (rocprofiler-compute)
- ROCm Systems Profiler (rocprofiler-systems)
- ROCprofiler-SDK
- ROCdbgapi
- ROCgdb
- ROCR Debug Agent

 Control and monitoring tools

- AMD SMI
- ROCm Data Center Tool
- rocminfo

 Media libraries

- rocDecode
- rocJPEG