To continue to enhance Radeon Open eCosystem, AMD has just released ROCm 5.0. As the company's open software platform, it is dedicated to giving researchers access to key AMD Instinct accelerator cards to drive more scientific discoveries. In addition, the Radeon ROCm platform was designed with portability in mind from the start, supporting proxy and architectural environments across numerous accelerators.
(From AMD official website)
With the release of Suzi and ROCm 5.0, AMD will facilitate code development for related solutions by bringing access to available computing languages, compilers, libraries, and tools for high-performance computing (HPC) and artificial intelligence (AI) research.
Expanding the Radeon open ecosystem with the Instinct MI200 Series accelerator cards and delivering unprecedented performance across critical workloads, ROCm 5.0 also introduces new support for Radeon Pro V620 / W6800 workstation GPUs.
What is ROCm - ExplainingTech（via）
RNDA 2 (and some Navi 2x GPUs) became one of the first graphics cards to be supported by this open-source computing software stack, and ROCm's support for Navi 1x/2x has been absent for some time, mainly in the area of AMD's workstation GPU product line.
Although AMD has not officially promoted ROCm applications for its own consumer graphics cards, in most cases, the Radeon RX 6000 series GPUs are expected to be used with ROCm due to the corresponding workstation (Radeon Pro) version.
ROCm support for Navi 1x is still relatively limited, but ideally, AMD can solve certain problems over time.
As for when the company intends to introduce support for Polaris GPUs, it is not known (ROCm 4.5 was already disappointed once when it was released).
GitHub (Portal)
Finally, the official release notes detail AMD ROCm 5.0, which improves the help and support for Red Hat Enterprise Linux 8.5 and introduces partial mode driver code for AMDGPU/AMDKFD.
There are further establishment/expulsion processing, more content on the docs.amd.com documentation portal, and other HIP updates to facilitate the transfer of the CUDA codebase to run on AMD GPUs.