Hi Christian, On 2023-07-03 08:31, Christian Kastner wrote:
We plan to expand ROCm support from the currently supported AMD RDNA 2 workstation GPUs: the Radeon Pro v620 and w6800 to select AMD RDNA 3 workstation and consumer GPUs. Formal support for RDNA 3-based GPUs on Linux is planned to begin rolling out this fall, starting with the 48GB Radeon PRO W7900 and the 24GB Radeon RX 7900 XTX, with additional cards and expanded capabilities to be released over time.
A couple notes:A. The Radeon PRO W7900, Radeon PRO W7800, Radeon RX 7900 XTX and Radeon RX 7900 XT are all Navi 31 (gfx1100). While only the W7900 and RX 7900 XTX will be officially supported by upstream, it's all just gfx1100 from the perspective of the compiler. I would be surprised if there are any issues on other Navi 31 GPUs despite not being officially supported.
B. In ROCm 5.4, the math and communication libraries were enabled and validated by their development teams on gfx1100. In ROCm 5.5, the math and communication libraries were enabled on gfx1102 (Navi 33; e.g., Radeon RX 7600). I think MIOpen got initial support for gfx1100 in ROCm 5.5 as well. In ROCm 5.6, I think MIOpen was enabled on gfx1102 and appears to have had additional tuning done for gfx1100 (based on the build architectures for Composable Kernel [4]).
C. The official support list for ROCm 5.6.0 [5] does not include RDNA 3 GPUs, as they have not been fully validated yet.
D. We can start building for GFX11 on Debian once we update HIP and move from LLVM 15 to LLVM 16. The LLVM 16 toolchain is already available on Unstable. However, we have a stable stack for GFX9 and GFX10 hardware and would like to wrap up my packaging of the test suites for the ROCm math libraries before we take on that transition.
Sincerely, Cory Bloor[4]: https://github.com/ROCmSoftwarePlatform/MIOpen/blob/rocm-5.6.0/requirements.txt#L9 [5]: https://rocm.docs.amd.com/en/docs-5.6.0/release/gpu_os_support.html#supported-gpus