In cuda11.0, running time become slower than cuda10.x

[JunHeum]

Hi,

I found that using CUDA11.0+PyTorch1.7.0+CUDNN8 almost doubles the running time(train and test) compared to using CUDA10.2+PyTorch1.7.0+CUDNN7.

To use RTX 30 series graphic cards, we must use CUDA11 and CUDNN 8.

When you afford to experiment, I am wondering if I could know what parts in your codes make this issue.

[zachteed]

Hi, I will look into this in the next few days

[gaoyumin11]

I would like to know if there is a better solution

[MiaoJieF]

I found out that this is due to the num worker number being set too small.