z-fabian/HUMUS-Net

CPU utilization is particularly high

Opened this issue · 0 comments

I found that CPU utilization was so high during training that my GPU was waiting for cpu processing for a long time. It made me wonder that another model (like the modified adjacent VarNet) did not have as high CPU utilization under the same data processing conditions, especially when most of the operations should be performed on the GPU. Is it the same when you train?
1703775182403