Question: If I have multiple GPUs, can I specify to run colabfold on separate GPU?

[JasonIsaac-Lofty]

I have multiple GPU on our server, can I run colabfold on separate GPU like alphafold2, just by using the flag: --gpu_devices?

[ChrisLou-bioinfo]

yes？

[sean-workman]

This is a really unhelpful answer. There is no --gpu_devices flag in the Flags section of the read me or in the usage message from --help.

Edit:

Editing to add that --gpu_devices is indeed not a recognized flag.

[YoshitakaMo]

How about setting CUDA_VISIBLE_DEVICES environment variable manually? For example,

CUDA_VISIBLE_DEVICES=0 colabfold_batch --amber --templates <input> <outputdir>
CUDA_VISIBLE_DEVICES=1 colabfold_batch --amber --templates <input> <outputdir>
...

[sean-workman]

Hi @YoshitakaMo, I did indeed end up setting specific devices as visible in the bash scripts I wrote. Thank for maintaining this great resource!

[JasonIsaac-Lofty]

How about setting CUDA_VISIBLE_DEVICES environment variable manually? For example,

CUDA_VISIBLE_DEVICES=0 colabfold_batch --amber --templates <input> <outputdir>
CUDA_VISIBLE_DEVICES=1 colabfold_batch --amber --templates <input> <outputdir>
...

Thank you very much! That was really helpful!