ROCm 7.0.2 released
ROCm 7.0.2 is a new update for the platform that adds many new features and makes a lot of things better, such as supported hardware, operating systems, and software libraries.
AMD ROCm is a set of open-source software that includes drivers, development tools, and APIs that make it easier to program GPUs from the kernel level to end-user applications. ROCm is a complete solution that is optimized for generative AI and high-performance computing (HPC) applications. It gives developers all the tools they need to make high-performance computing solutions.
One big improvement is that the RDNA4 architecture-based AMD Radeon RX 9060 is now supported, which means that more hardware configurations can work with it. The release also makes it compatible with newer operating systems like Debian GNU/Linux 13 (Trixie), Oracle Linux 10, and Red Hat Enterprise Linux (RHEL) 10. For people who use AMD Datacenter GPU products, it is essential to keep an interdependent stack of user space, driver, and firmware for the best performance.
One of the most important things about this release is that the AMD GPU Driver (amdgpu) can now reset multimedia engines better, especially for MI300X GPUs. This update greatly improves the stability of the system and the resilience of the GPU by making it easier to deal with possible errors. Users of RHEL 10.0 or Oracle 10 must make sure they have the most recent version of the AMD GPU Driver to make the most of these improvements.
With the addition of Retrieval-Augmented Generation (RAG) support, developers can build accurate AI pipelines that use real-time data. Gaussian splatting (gsplat) support, on the other hand, speeds up 3D rasterization on AMD Instinct GPUs using an open-source library. This integration is likely to have a big effect on many applications, such as computer-aided design and medical imaging.
ROCm Life Science (ROCm-LS) is a new toolkit that was released but is still in early access. It has a lot of potential to speed up life science applications that use GPU acceleration. Researchers could find the toolkit very useful because it comes with important libraries like hipCIM for image processing and MONAI for high-performance medical AI applications.
The addition of new support for PyTorch 2.8 and the release of FlashInfer and llama.cpp frameworks for large language model applications have greatly improved the development of deep learning and AI. These improvements should help people who work on advanced projects in fields like natural language processing and computer vision.
The installation tools, such as the ROCm Offline Installer Creator and Runfile Installer, have been updated to better support making offline installers that are specific to certain distributions. Furthermore, the ROCm documentation has been updated with new tutorials aimed at AI developers. The goal is to make it easier for people to use and adopt the platform.
ROCm is easy to use and move from the existing code, which is one of its best features. Because the software stack is open source, developers can easily add their apps to ROCm and take advantage of the optimized performance of AMD GPUs. This makes it a good choice for developers who want to move their apps to high-performance computing environments.
Updates to component versions show that libraries are constantly getting better in terms of functionality, compatibility, and user experience. This is especially true in areas like fixing bugs and making things run faster. However, it's important to note that there are known problems with ROCm tools on distributions that have SELinux enabled. These problems could make GPU code debugging unresponsive.
The ROCm team is going to stop supporting some parts, like ROCm Execution Provider, ROCm SMI, and some profiling tools, to get ready for future releases. Users should introduce changes to their projects as new options become available to make the transition as smooth as possible.
ROCm 7.0.2 Release
ROCm 7.0.2 release notes The release notes provide a summary of notable changes since the previous ROCm release.
