Only 1 CPU is used when training, batch size 20, num_workers 20

[davodogster]

Is this normal? It seems like I am wasting CPU resources

[rusty1s]

The DataLoader should use 20 workers, and the GNN should use multi-threading for heavy computations as well (except for scatter_* ops which are single-threaded). If you are only using 1 CPU with num_workers=20, I assume there is something wrong with your PyTorch installation?

[davodogster]

Do you thinks it's because I'm training with InMemory?

[rusty1s]

This shouldn't be related AFAIK.