GPU mode only supports one CPU thread #122
Description
Hi,
Many loves to your toolbox. Our dataset has different contrast settings than the regular MP-RAGE images, and your model is a good fit.
I am using the py 3.8 version in docker on WSL2. My GPU is RTX 4070, and I believe it is overqualified for this unet. However, my prediction took almost 16 minutes for one .nii with the --fast flag. Then I tried to monitor my CPU and GPU utilization during running and figured out there was only one CPU thread be called in whatever GPU mode (even though I indicated a specific number to the --threads flag). The CPU processing heavily delayed my time.
However, once I switched to the --cpu mode and indicate --threads (as 20 in this case). The job is done within 30 seconds.
