gpu requirements

[voa18105]

gpu requirements

[rosinality]

I have used P40 with 24G ram. Maybe it is better to use more gpus with dataparallel.

[voa18105]

I have 2 TITAN X with 12G, but use only 1... is there a simple way to insert data parallelization?

[voa18105]

ho, I see you already used nn.DataParallel... Does that mean that if I will write CUDA_VISIBLE_DEVICES=0,1 python3.6 train.py -d celeba ... it will automatically use both?

[rosinality]

Yes, it will use both gpus. I think it would be better to test that you can train model at high resolutions. (You can do this by manually setting step variables.)

[voa18105]

Thank you! I will notify in 1-2 days if I managed to overcome this issue.

[RameenAbdal]

Hi, Did you manage to run the code on multiple gpus?

[voa18105]

@RameenAbdal hi, yes, it works fine. I guess it is much slower than with P40, but it does not fall