How to use PyTorch with ROCm (radeon gpu)? How to transfer data to gpu?

[oconnor127]

Hey,
So far I didnt see any documentation or similar, which gives a hint how to use PyTorch with other GPUs than NVIDIA (when the new ROCm package is installed). How can I choose my radeon GPU as device and so use it for training? Very glad for any advices.

Best

cc @jeffdaily @sunway513 @jithunnair-amd @ROCmSupport

[imaginary-person]

The ROCm version is used in the same way as the CUDA version:
eg. t = torch.tensor([5, 5, 5], dtype=torch.int64, device='cuda')

[oconnor127]

Thank you very much. Guess my Radeon RX580 is not supported yet. Any rough date, when you think the newer GPU will be supported as well?

[imaginary-person]

Guess my Radeon RX580 is not supported yet.

I think AMD ROCm doesn't officially support it anymore, but this link also states, Some of this software may work with more GPUs than the "officially supported" list above, though AMD does not make any official claims of support for these devices on the ROCm software platform. So maybe the AMD folks CCed in this issue can clarify.

[imaginary-person]

@oconnor127, you might want to take a look at #53738 & check if the corresponding workaround exists for your GPU.

[damico]

I just made a python code that tests PyTorch support for AMD GPUs with ROCM: https://gist.github.com/damico/484f7b0a148a0c5f707054cf9c0a0533

[hongxiayang]

@oconnor127 This is the updated documentation about what GPUs are supported: https://rocm.docs.amd.com/en/latest/release/gpu_os_support.html

We are going to close this issue as this is very old.