AMD has released ROCm 10.0, a major update to its open-source Radeon/Instinct GPU compute stack. The release addresses previous versioning inconsistencies and introduces ROCm.AI for AI-driven development, along with a new command-line interface (CLI).
ROCm 7.14 was previously announced as a production release, built on TheRock build system and supporting Ryzen AI 400 series. However, the versioning was unusual given the existing ROCm 7.2 stable release and the “tech preview” status of ROCm 7.9+. AMD has now moved to ROCm 10.0 to resolve this, succeeding the ROCm 7.x series.
Key features of ROCm 10.0 include ROCm.AI and the ROCm CLI. The release also marks AMD’s shift to a six-week release cycle. Hyperloom, announced at AMD’s Advancing AI event, is also included. Improvements have been made to Linux package consolidation, with production-ready containers and Python wheels for vLLM, and local LLM fine-tuning for Ryzen AI Max.
For Windows users, AMD has retired the HIP SDK in favor of the ROCm Core SDK. Windows will maintain the same release cadence as Linux, with the current ROCm Core SDK being a static package, and native installer support planned for later this year. Further details are available on the AMD ROCm blog.
Read the original coverage
💬 Comments
📜 Comment Policy