If I have both AMD and NVIDIA? #16557
Replies: 2 comments 3 replies
-
the intended way for you to use whichever pytorch you need is for you to set the environment variable stable-diffusion-webui/webui-user.sh Lines 27 to 28 in 82a973c TORCH_COMMAND contains the command that we execute when installing pytorch
if |
Beta Was this translation helpful? Give feedback.
-
If you don't need it, you can usually disable integrated graphics in the motherboard BIOS. |
Beta Was this translation helpful? Give feedback.
-
Hi folks!
Yesterday I've assembled an old computer at home specifically to use with SD (and maybe other AIs).
When it came to installing AUTOMATIC1111 I've noticed that it downloads some rocm version of torch and then gives me an error of something like "couldn't find a version that satisfies..." Anyway, Googling what this rocm is in the first place, gave me an answer that it's a thing for AMD cards... But that made me scratch my head like "Why?! I clearly have an NVIDIA card, driver installed, nvidia-smi shows it's info, everything's fine!"
But then I've remembered that this motherboard has an AMD card as an integrated video! So I've opened webui.sh and found-out that it greps from lspci output and if it's an AMD - downloads rocm packs and if NVIDIA - then cuda ones. And lspci on this machine indeed shows AMD card before NVIDIA in a list of devices.
So I've just commented-out the parts where it checks for AMD and downloads rocm packages. That seemed to do the trick and it started downloading torch??cu*something.. But it was already like 3a.m. and I've decided to continue this process tomorrow, lol.
So my question is - is there some more elegant way of archiving this other then "surgically" removing parts of the script?
Maybe there's some key to force CUDA or something?
Beta Was this translation helpful? Give feedback.
All reactions