Skip to main content
Ctrl+K
AMD Logo
ROCm™ Software 6.4.1 Version List
  • GitHub
  • Community
  • Blogs
  • ROCm Developer Hub
  • Instinct™ Docs
  • Infinity Hub
  • Support

ROCm Documentation

  • What is ROCm?
  • Release notes
  • Compatibility matrix
    • Linux system requirements
    • Windows system requirements

Install

  • ROCm on Linux
  • HIP SDK on Windows
  • ROCm on Radeon GPUs
  • Deep learning frameworks
  • Build ROCm from source

How to

  • Use ROCm for AI
    • Installation
    • System health benchmarks
    • Training
      • Train a model with Megatron-LM
      • Train a model with PyTorch
      • Train a model with JAX MaxText
      • Train a model with LLM Foundry
      • Scale model training
    • Fine-tuning LLMs
      • Conceptual overview
      • Fine-tuning
        • Use a single accelerator
        • Use multiple accelerators
    • Inference
      • Run models from Hugging Face
      • LLM inference frameworks
      • vLLM inference performance testing
      • PyTorch inference performance testing
      • Deploy your model
    • Inference optimization
      • Model quantization techniques
      • Model acceleration libraries
      • Optimize with Composable Kernel
      • Optimize Triton kernels
      • Profile and debug
      • Workload optimization
    • AI tutorials
  • Use ROCm for HPC
  • System optimization
  • AMD Instinct MI300X performance guides
  • System debugging
  • Use advanced compiler features
    • ROCm compiler infrastructure
    • Use AddressSanitizer
    • OpenMP support
  • Set the number of CUs
  • Troubleshoot BAR access limitation
  • ROCm examples

Conceptual

  • GPU architecture overview
    • MI300 microarchitecture
      • AMD Instinct MI300/CDNA3 ISA
      • White paper
      • MI300 and MI200 Performance counter
    • MI250 microarchitecture
      • AMD Instinct MI200/CDNA2 ISA
      • White paper
    • MI100 microarchitecture
      • AMD Instinct MI100/CDNA1 ISA
      • White paper
  • File structure (Linux FHS)
  • GPU isolation techniques
  • Using CMake
  • Inception v3 with PyTorch

Reference

  • ROCm libraries
  • ROCm tools, compilers, and runtimes
  • Accelerator and GPU hardware specifications
  • Hardware atomics operation support
  • Precision support
  • Graph safe support

Contribute

  • Contributing to the ROCm documentation
    • ROCm documentation toolchain
    • Building documentation
  • Providing feedback about the ROCm documentation
  • ROCm licenses
  • ROCm libraries

ROCm libraries

ROCm libraries#

2025-05-22

2 min read time

Applies to Linux and Windows

Machine Learning and Computer Vision
  • Composable Kernel

  • MIGraphX

  • MIOpen

  • MIVisionX

  • rocAL

  • rocDecode

  • rocPyDecode

  • rocJPEG

  • ROCm Performance Primitives (RPP)

Primitives
  • hipCUB

  • hipTensor

  • rocPRIM

  • rocThrust

Communication
  • RCCL

  • rocSHMEM

Math
  • half

  • hipBLAS / rocBLAS

  • hipBLASLt

  • hipFFT / rocFFT

  • hipfort

  • hipRAND / rocRAND

  • hipSOLVER / rocSOLVER

  • hipSPARSE / rocSPARSE

  • hipSPARSELt

  • rocALUTION

  • rocWMMA

  • Tensile

previous

Deep learning: Inception V3 with PyTorch

next

ROCm tools, compilers, and runtimes

  • Terms and Conditions
  • ROCm Licenses and Disclaimers
  • Privacy
  • Trademarks
  • Supply Chain Transparency
  • Fair and Open Competition
  • UK Tax Strategy
  • Cookie Policy
  • Cookie Settings
© 2025 Advanced Micro Devices, Inc