How to assign a different GPU?

[TangYiChing]

Hi,

Due to the GPU-Out-Of-Memory problem, I have to change the default "cuda:0" to another empty GPU, Here's what I did to set the GPU device by changing cuda:0 to cuda:1:

vim igfold/IgFoldRunner.py 

device = torch.device(
"cuda:0" if torch.cuda.is_available() and try_gpu else "cpu")  
print(f"Using device: {device}")

However, I got this error: RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:1 and cuda:0! (when checking argument for argument index in method wrapper_CUDA__index_select)

Can you tell me how to assign a GPU appropriately? Thank you.

[aozalevsky]

@TangYiChing the easy hotfix is to:

1. change the line

https://github.com/Graylab/IgFold/blob/85d9dcdbc7afb9c38c4427df656fc194a61bd6ee/igfold/IgFoldRunner.py#L61C35-L61C35

from cuda:1 to just cuda

2. select just one cuda device with a few additional lines:

   import os
   os.enivron["CUDA_VISIBLE_DEVICES"]="0"

Here is an explanation of how this variable works.