Closed Asquator closed 10 months ago
cgmb
commented
1 year ago Could it be that my integrated GPU is not supported by ROCm/rocblas?
That is correct. There are no integrated GPUs that are officially supported.
UPD: Setting local variables recommended in similar issues won't help.
You mean you tried running after setting export HSA_OVERRIDE_GFX_VERSION=9.0.0 and it didn't work? That's really the only thing I could think of to try.
daineAMD
commented
1 year ago HI @Asquator, thanks for raising this issue.
You can see the list of supported GPUs in the ROCm 5.6 Documentation. As Cory mentioned you can try setting HSA_OVERRIDE_GFX_VERSION=9.0.0.
If you have any other questions feel free to ask, otherwise I'll close this issue as unsupported.
Thanks, Daine
EXtremeExploit
commented
1 year ago i am having this exact same problem, rx 580 on arch linux, setting HSA_OVERRIDE_GFX_VERSION didnt help (sometimes its illegal seek and others its no such file)
daineAMD
commented
10 months ago Hi again, As mentioned, you can see the list of officially supported GPUs in the ROCm Documentation. The integrated graphics in the original issue and rx580 (gfx803) are both unsupported.
I'll close this issue, feel free to reach out if you have further questions. Thanks, Daine
Describe the bug
The notorious error pops up when trying to send a pytorch tensor to GPU:
rocBLAS error: Cannot read /home/<PATH>/venv/lib/python3.11/site-packages/torch/lib/rocblas/library/TensileLibrary.dat: Illegal seekTo Reproduce
Try to train a pytorch model on GPU.
Expected behavior
Successful work with pytorch tensors on GPU.
Environment
Additional context
Will happily provide any logs if I'm able to find them. The execution simply fires the error and freezes, no other messages are displayed. Could it be that my integrated GPU is not supported by ROCm/rocblas?
UPD: Setting local variables recommended in similar issues won't help.