Closed raywzy closed 5 years ago
Hi @raywzy , Thanks! I have very little experience with multi-GPU configurations, but GeomLoss should run smoothly with the standard PyTorch interface:
gpuids = [0,1] if torch.cuda.device_count() > 1 else [0]
for gpuid in gpuids:
with torch.cuda.device(gpuid):
# Create your data on the gpu with .cuda(), work with GeomLoss, etc.
There is a short tutorial on the subject in the KeOps documentation. I cannot try it right now, but would expect GeomLoss to handle this transparently. Does it work on your machine?
Best regards, Jean
@jeanfeydy That works fine for me. Thanks a lot :D
You're welcome :-)
Very great package! My question is how to set the gpu device?