fairy-stockfish / variant-nnue-pytorch

chess variant NNUE training code for Fairy-Stockfish
https://github.com/fairy-stockfish/variant-nnue-pytorch/wiki/Introduction
28 stars 18 forks source link

Multi-GPU training #15

Closed Newera2022 closed 1 year ago

Newera2022 commented 2 years ago

Thanks

ianfab commented 2 years ago

I unfortunately do not have a multi-GPU setup where I could test this, so it is hard for me to tell. What happens when you pass something like --gpus 2? Does it fail or just not use more than one GPU?

ianfab commented 2 years ago

What is the error message you get? And how many GPUs do you have?

ianfab commented 2 years ago

Ok, thanks. To my knowledge instead of the count you can also list the ids of the specific GPUs you would like to use, e.g., --gpus "0,1" or the like, not sure if that might work.

ianfab commented 2 years ago

It looks like multiple GPU training isn't supported by the current code. I don't have any multi GPU machines, and normally anyway the training data generation is the bottleneck, so I never looked into that so far.