Installing for rocm on a system where no NVIDIA gpu is discovered but AMD is discovered always installs CUDA
Stability Matrix has a setting for preferred pytorch backend. No matter what is selected for ComfyUI, AIII, or others, it always installs the nvidia CUDA version. When explicitly selecting rocm, it installs CUDA. There is a lot of code to detect AMD && !NVIDIA and install rocm, but none of that seems to do anything in 2.6.7 at least.
Similar problem, after installing any package with ROCM version of PyTorch selected I get message about NVIDIA driver not installed.
the AMD card shows up in settings, rocminfo also returns the card
version 2.11.4
Having this same issue, best i can get out of any interface is cpu generation. Can't get anything to actually use my 6800xt
Edit: finally got automatic1111 to use rocom but comfyUI and others refuse always complaining about no Nvidia even when i explicitly select rocom when installing
I have this same issue. Only way I can use Comfy is if I manually uninstall the pytorch packages and reinstall the ones for rocm. :/
same issue here
This issue is stale because it has been open 30 days with no activity. Remove the stale label or comment, else this will be closed in 5 days.
despise deep developer wishes, bug do not disappear when you stop looking at them. Go away @github-actions bot
This issue is stale because it has been open 30 days with no activity. Remove the stale label or comment, else this will be closed in 5 days.
This issue was closed because it has been stale for 5 days with no activity.