AMD has released ROCm 10.0, a major update to its open-source GPU compute stack, designed to support agentic AI applications. This release marks a decade since the initial launch of ROCm 1.0 in 2016. The new version is built on TheRock, an automated open-source build and release system that reached production in ROCm 7.14. The update introduces a native agentic AI developer experience called ROCm.AI, which integrates the ROCm CLI, AMD Skills, and Hyperloom. This new experience streamlines the workflow for developers working on AI workloads.
ROCm 10.0 is the first major version bump since the 7.x series and is built end-to-end on TheRock. This system allows for a single pipeline to generate primitives, libraries, and framework wheels, which are then validated across the entire stack before release. The release cycle for ROCm 10 and its minor versions is approximately every six weeks, ensuring consistent updates across AMD's GPU portfolio, including Instinct accelerators, Radeon graphics cards, and Ryzen integrated graphics. The unified platform ensures that the same stack is used for both prototyping on laptops and deploying in data centers.
The release also consolidates GPU software distribution onto a redesigned repo.amd.com. This change simplifies the process of obtaining a full GPU stack by bringing all necessary components—ROCm packages, the amdgpu driver, and AMD’s public GPU tools—into a single repository. The new layout supports multiple ROCm versions and architectures, allowing teams to manage versions and roll out updates independently of the release train. This change is particularly beneficial for automation in installation processes, as it reduces the complexity of managing multiple repositories.
Source: amd
