faststyle
faststyle copied to clipboard
GPU scaling
Is there any way to train with multiple GPUs? The current scripts seem to only utilize the first GPU, altough TF 'sees' all of them.
Unfortunately, I haven't yet played with the multi-tower training stuff as I only have a humble little 970 at the moment. Pull requests are welcomed!
ok, thanks. I am quite new to TF though - will take a look.
I'll leave this issue open in case someone implemented it in one of their forks and sees this.