AMD has announced the release of ROCm 7.1.1, an updated version of its open-source software stack tailored for GPU computation. This latest version enhances the ecosystem for developers by allowing customization of GPU software, targeting high-performance computing (HPC), artificial intelligence (AI), scientific computing, and computer-aided design (CAD).
The update focuses on practical improvements, particularly for developers utilizing AMD's Instinct GPUs. It enhances performance in areas such as HPC and AI research, making it easier to achieve reliable results. Notable upgrades include improved support for virtualization, specifically KVM SR-IOV, with new guest operating system options such as Ubuntu 24.04 and support for RHEL versions 9.7 and 10.1.
Developers using MI350X or MI355X GPUs will benefit from a more efficient SMI tool that reduces repetitive tasks through caching, yielding more detailed performance statistics. Additionally, improvements have been made to address previous issues with MI325X cards in SR-IOV Mode 1 and enhance GEMM kernel selection efficiency for matrix operations.
For deep learning practitioners, ROCm 7.1.1 includes compatibility with PyTorch version 2.9, strengthening the ROCm ecosystem. Hugging Face Transformers models can now be developed using the gfx1201 GPU architecture. A significant milestone is the ROCm Data Science Toolkit, which has reached general availability status, with components hipDF and hipMM now fully production-ready. Two new tools, hipRAFT and hipVS, have also made their debut.
Performance enhancements extend to DGL (for graph neural networks) and llama.cpp, alongside improvements for creating offline installers. A range of fixes has been implemented in various areas of ROCm, including AMD SMI, Composable Kernel setup, and machine learning tools such as MIGraphX and RCCL.
Looking ahead, the Composable Kernel component is set to adopt C++20 features in future updates, necessitating that developers prepare their build environments accordingly to avoid potential issues during upgrades.
In summary, ROCm 7.1.1 represents a significant advancement in AMD's open-source GPU platform, offering developers enhanced tools and functionalities to meet the demands of modern computing environments. This release not only addresses existing challenges but also lays the groundwork for future developments, making it a pivotal moment for AMD's commitment to high-performance computing and GPU software solutions
The update focuses on practical improvements, particularly for developers utilizing AMD's Instinct GPUs. It enhances performance in areas such as HPC and AI research, making it easier to achieve reliable results. Notable upgrades include improved support for virtualization, specifically KVM SR-IOV, with new guest operating system options such as Ubuntu 24.04 and support for RHEL versions 9.7 and 10.1.
Developers using MI350X or MI355X GPUs will benefit from a more efficient SMI tool that reduces repetitive tasks through caching, yielding more detailed performance statistics. Additionally, improvements have been made to address previous issues with MI325X cards in SR-IOV Mode 1 and enhance GEMM kernel selection efficiency for matrix operations.
For deep learning practitioners, ROCm 7.1.1 includes compatibility with PyTorch version 2.9, strengthening the ROCm ecosystem. Hugging Face Transformers models can now be developed using the gfx1201 GPU architecture. A significant milestone is the ROCm Data Science Toolkit, which has reached general availability status, with components hipDF and hipMM now fully production-ready. Two new tools, hipRAFT and hipVS, have also made their debut.
Performance enhancements extend to DGL (for graph neural networks) and llama.cpp, alongside improvements for creating offline installers. A range of fixes has been implemented in various areas of ROCm, including AMD SMI, Composable Kernel setup, and machine learning tools such as MIGraphX and RCCL.
Looking ahead, the Composable Kernel component is set to adopt C++20 features in future updates, necessitating that developers prepare their build environments accordingly to avoid potential issues during upgrades.
In summary, ROCm 7.1.1 represents a significant advancement in AMD's open-source GPU platform, offering developers enhanced tools and functionalities to meet the demands of modern computing environments. This release not only addresses existing challenges but also lays the groundwork for future developments, making it a pivotal moment for AMD's commitment to high-performance computing and GPU software solutions
ROCm 7.1.1 released
AMD has released ROCm 7.1.1, an update to its open-source software stack designed for graphics processing unit (GPU) computation. This release provides a comprehensive ecosystem for developers to customize their GPU software and achieve high-performance computing (HPC), artificial intelligence (AI), and scientific computing needs.
