Hi,
My question is related to loading input data using the combination of Dataset
and DataLoader
. Since pin_memory
option in a DataLoader
works for CPU tensors, I understand that there are two ways to load the input data:
- In the
Dataset
create CPU tensors and then usingpin_memory
transfer it to GPU - Directly create CUDA tensors in the
Dataset
and don’t usepin_memory
Which method is more efficient wrt data loading time and why? Any references would be helpful.
Thanks!