AMD this evening announced ROCm-DS as a new toolkit for their ROCm compute stack to accelerate data processing workloads on Instinct accelerators.
AMD ROCm-DS is short for ROCm Data Science and is intended to help expand the scope of data processing abilities on Instinct GPUs. ROCm-DS aims to make data-intensive applications more suitable for use on AMD GPUs/accelerators and in turn is based on the RAPIDS software… Yes, RAPIDS was started by NVIDIA for GPU-accelerated data science while now AMD is seeking to leverage that open-source RAPIDS code started by NVIDIA.
ROCm-DS is considered “early access” for the moment with additional libraries expected in H2’2025. ROCm-DS will be focused on accelerated data science, scalability across datasets and workloads, wide integration and compatibility, ease of use with software libraries like Pandas, and working with other software at large.
Those wishing to learn more about the newly-announced ROCm-DS can do so via the ROCm blog.
Separately, AMD also announced today AOMP 21.0-1 as the newest version of their LLVM/Clang compiler downstream focused on OpenMP offloading to Instinct/Radeon hardware. The new AOMP release re-bases against the latest ROCm 6.4 sources and have a variety of other improvements/fixes. Some of the other ROCm open-source components have also been seeing ROCm v6.4.1 version tags made today too in preparing for the next ROCm point release.
