You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Jun 22, 2024. It is now read-only.
I was wondering if we could add support for AMD GPUs running ROCm in the near future? I could help and could test it on two different AMD cards.
I think in order to get there, we might have to do the following:
Provide all the ROCm related packages.
Probably best to do this in another environment than the default one right now. I was wondering how we avoid duplicating most of the code here? (My Nix greatness is rather limited I have to say)
Set some environment variables, so that ROCm knows which GPU architecture we are using. This is particularly important in order to get this running on GPUs that are not officially supported by ROCm, but work nonetheless (which is basically all RDNA3, RDNA2 consumer and workstation GPUs, and also a significant number of RDNA-based GPUs).
Since users will have different GPUs, is there a (possibly hacky) way to pick this up dynamically from the environment?
The text was updated successfully, but these errors were encountered:
I was wondering if we could add support for AMD GPUs running ROCm in the near future? I could help and could test it on two different AMD cards.
I think in order to get there, we might have to do the following:
Provide all the ROCm related packages.
Probably best to do this in another environment than the default one right now. I was wondering how we avoid duplicating most of the code here? (My Nix greatness is rather limited I have to say)
Set some environment variables, so that ROCm knows which GPU architecture we are using. This is particularly important in order to get this running on GPUs that are not officially supported by ROCm, but work nonetheless (which is basically all RDNA3, RDNA2 consumer and workstation GPUs, and also a significant number of RDNA-based GPUs).
Since users will have different GPUs, is there a (possibly hacky) way to pick this up dynamically from the environment?
The text was updated successfully, but these errors were encountered: