search

nerdyrodent / VQGAN-CLIP

Just playing with getting VQGAN+CLIP running locally, rather than having to use colab.
Other
2.61k stars 428 forks source link

Accidental multi-GPU? #47

Open HughPH opened 3 years ago

HughPH commented 3 years ago

I have a cut of this code from a week or two ago.

Funnily enough I also added the option to run it on another GPU. When I do choose cuda:1, though, I get 2GB allocated on cuda:0 although that device is not specified anywhere in generate.py. Combined with disabling ECC (nvidia-smi -i 1 -e 0) this is Fine, because I can get over 912KibiPixels (1280x720 or 1488x624), but it would be good to understand what, why and how.

rlallen-nps commented 3 years ago

I have a cut of this code from a week or two ago.

Funnily enough I also added the option to run it on another GPU. When I do choose cuda:1, though, I get 2GB allocated on cuda:0 although that device is not specified anywhere in generate.py. Combined with disabling ECC (nvidia-smi -i 1 -e 0) this is Fine, because I can get over 912KibiPixels (1280x720 or 1488x624), but it would be good to understand what, why and how.

When reading up on torch's dataparallel I saw several mention a similar issue. May want to start there.