How fast should training be? Getting 2.3 steps/s on RTX 3080
serg06 opened this issue · comments
It looks like my GPU's only at 50% usage:
And the speed of python train_tacotron.py is only 2.3 steps/s:
Even though I have an RTX 3080 and a Ryzen 3950X. Is that normal?
Update: Looks like it might be because PyTorch and cuDNN aren't optimized for 3000-series yet.
Closing for now, will re-open if needed afterwards.