ROCm 7.1.0 released
The latest iteration of AMD's open-source software stack, ROCm 7.1.0, officially releases an array of tools and features that help developers optimize high-performance computing (HPC) applications, particularly those focused on generative AI. This comprehensive solution encompasses drivers, development tools, and APIs, allowing for seamless GPU programming from the kernel level up to end-user applications.
ROCm 7.1.0 boasts a multitude of significant updates and enhancements that are designed to amplify functionality on AMD Radeon GPUs and Ryzen APUs. A notable expansion in operating system support has been achieved, with AMD Instinct MI325X and MI100 GPUs now able to function seamlessly with RHEL 10.0, SLES 15 SP7, Debian 13 and 12, and Oracle Linux 9 and 10. This increased compatibility ensures that developers can leverage the full potential of ROCm across various platforms.
Among the key features introduced in this release is the ability to set a power cap for AMD Instinct MI300X in virtualized environments, which significantly improves overall system efficiency and reduces energy consumption. Additionally, better support for multiple users has been added for MI350 Series GPUs, and the HIP (Heterogeneous-Compute Interface for Portability) runtime has received several updates to align more closely with NVIDIA CUDA standards.
Several improvements have been made to different libraries, such as better performance in hipBLASLt for AMD Instinct GPUs and enhanced SpMM features in hipSPARSELt. Furthermore, new functionalities have been added to rocAL for Vision Transformer model training, making ROCm an even more potent tool for developers tackling complex deep learning tasks.
Deep learning frameworks also stand to benefit from this release, with extended support now available for TensorFlow 2.20.0 and the adoption of MIGraphX for more efficient inference tasks. This streamlined approach enables developers to focus on creating innovative AI solutions without being bogged down by infrastructure complexities.
Other important updates include better profiling tools like the ROCm Compute Profiler and ROCm Systems Profiler, which now collect and analyze data more effectively, enabling real-time monitoring and the ability to attach to processes on the fly. Additionally, the end of support for the ROCm Execution Provider (ROCm-EP) has been announced, with users being transitioned to the MIGraphX Execution Provider for better compatibility and support.
Documentation enhancements have also been made to provide clearer guidance for users, featuring new tutorials added for AI developers and updates in the HIP documentation to support streamlined GPU application transformation. Performance improvements have been implemented across ROCm components, ensuring greater efficiency in operations, especially for diverse machine learning and AI workloads.
ROCm/ROCm Release ROCm 7.1.0 Release
ROCm 7.1.0 release notes The release notes provide a summary of notable changes since the previous ROCm release.
