ROCm 10.0.0 Support for Ray 2.58.0: Enabling Scalable AI Workflows
AMD, Tuesday, October 27th, 2026
AMD details ROCm 10.0.0 support for Ray 2.58.0, letting existing Ray code run unchanged on AMD Instinct GPUs.
The AMD ROCm blog announces Ray 2.58.0 support on ROCm 10.0.0, with a validated stack including PyTorch 2.12, vLLM on ROCm, and current Instinct GPUs such as MI355X.
Ray discovers Instinct GPUs automatically and recognizes MI300X, MI325X, and MI355X as native accelerator types that workloads can pin, which the author recommends mainly for mixed-SKU or mixed-vendor clusters.
The article walks through installing Ray on ROCm, verifying GPU visibility, and running a Ray Serve LLM example that serves a tensor-parallel vLLM model behind an OpenAI-compatible endpoint.
It also points to verl for RLHF workloads.