Use ROCm on Radeon GPUs

Use ROCm on Radeon GPUs#

Turn your desktop into a Machine Learning platform with the latest high-end AMD Radeon™ 9000 and 7000 series GPUs

AMD has expanded support for Machine Learning Development on RDNA™ 4 and RDNA™ 3 GPUs with Radeon™ Software for Linux 25.10.1 with ROCm™ 6.4.1

Researchers and developers working with Machine Learning (ML) models and algorithms using PyTorch, ONNX Runtime, JAX, or TensorFlow can now also use ROCm 6.4.1 on Linux® to tap into the parallel computing power of the newest high-end AMD Radeon GPUs for desktop, including:

  • AMD Radeon™ AI PRO R9700

  • AMD Radeon™ RX 9070

  • AMD Radeon™ RX 9070 XT

  • AMD Radeon™ RX 9070 GRE

  • AMD Radeon™ RX 9060 XT

These new GPUs based on the RDNA 4 architecture join the already-supported Radeon 7000 series built on RDNA 3, further expanding support for high-performance local ML development on Linux®.

A client solution built on powerful high-end AMD GPUs enables a local, private, and cost-effective workflow to develop ROCm and train Machine Learning for users who were solely reliant on cloud-based solutions.

More ML performance for your desktop

  • With today’s models easily exceeding the capabilities of standard hardware and software not designed for AI, ML engineers are looking for cost-effective solutions to develop and train their ML-powered applications. Due to the availability of significantly large GPU memory sizes of 24GB or 48GB, utilization of a local PC or workstation equipped with the latest high-end AMD Radeon 9000 & 7000 series GPU offers a robust/potent yet economical option to meet these expanding ML workflow challenges.

  • Latest high-end AMD Radeon 9000 series GPUs are built on the RDNA 4 GPU architecture and AMD Radeon 7000 series GPUs are built on the RDNA 3 GPU architecture

    • offers up to 24GB or 48GB of GPU memory to handle large ML models

Migrate your application from the desktop to the datacenter

  • ROCm is the open-source software stack for Graphics Processing Unit (GPU) programming. ROCm spans several domains: General-Purpose computing on GPUs (GPGPU), High Performance Computing (HPC) and heterogeneous computing.

  • The latest AMD ROCm 6.4.1 software stack for GPU programming unlocks the massively parallel compute power of these RDNA 4 and RDNA3 GPUs for use with various ML frameworks. The same software stack also supports AMD CDNA™ GPU architecture, so developers can migrate applications from their preferred framework into the datacenter.

Freedom to customize

ROCm is primarily Open-Source Software (OSS) that allows developers the freedom to customize and tailor their GPU software for their own needs while collaborating with a community of other developers, and helping each other find solutions in an agile, flexible, rapid and secure manner. AMD ROCm allows users to maximize their GPU hardware investment. ROCm is designed to help develop, test and deploy GPU accelerated HPC, AI, scientific computing, CAD, and other applications in a free, open-source, integrated and secure software ecosystem.

Improved interoperability

  • Support for PyTorch, one of the leading ML frameworks.

  • Support for ONNX Runtime to perform inference on a wider range of source data, including INT8 with MIGraphX.

  • Support for TensorFlow.

Radeon™ Software for Linux® 25.10.1 with ROCm 6.4.1 Highlights

  • Support for Llama.cpp

  • Forward Attention 2 (FA2) backward pass enablement

  • Support for JAX (inference)

  • New models
    • Llama 3.1

    • Qwen 1.5

    • ChatGLM 2/4

  • Support for RedHat Enterprise Linux (RHEL) 9.6

Note

Visit AMD ROCm Documentation for the latest on ROCm.

For the latest driver installation packages, visit Linux Drivers for Radeon Software.