AMD looks to have launched ROCm 5.0, which is designed to improve the Radeon Open Ecosystem. ROCm is AMD’s open software platform, which allows researchers to use AMD Instinct accelerators to accelerate scientific discoveries. The Radeon ROCm platform was built with portability in mind, enabling scenarios across a wide range of accelerator agents and architectures.
With AMD Instinct MI200 series accelerators, AMD expands its platform for leading HPC and AI applications, extending support and accessibility of ROCm for developers and delivering unmatched performance across essential workloads. AMD’s RDNA2-based graphics cards are the first to be supported by the open-source computing software stack for graphic processors, with ROCm 5.0 adding support for Radeon Pro V620 and Radeon Pro W6800 workstation graphics cards. AMD has released a new platform update that includes support for some Navi 2x and RDNA2 GPU areas.
Support for Navi 1x/2x has been absent from ROCm for a long time. Some Navi components were working in ROCm 4.5, but there is now official Navi 2x support for their workstation GPUs in ROCm 5.0. Although AMD does not publicly advocate ROCm for consumer-level graphics cards, the Radeon RX 6000 series will most likely operate well with ROCm now that workstation support has been published.
In ROCm, the Navi 1x support has been inadequate; however, ROCm 5.0 should, in theory, fix some of these issues over time. In addition, it remains to be seen whether the Polaris GPU functionality in ROCm 5.0 has been restored. The Polaris support was failing with ROCm 4.5. They indicated that it was an unintentional relapse by still latently supporting those more experienced graphics processors; nevertheless, there is no indication in their most recent release information that Polaris support is on the way back.
The release notes go into greater information about AMD ROCm 5.0, including support for Red Hat Enterprise Linux 8.5 and the AMDGPU/AMDKFD part mode driver code from ROCm. In addition, the notes reveal improved establishment/expulsion handling, a new documentation gateway at docs.amd.com, and other HIP changes aimed at making it easier to port CUDA code to AMD GPUs.
also read:
