I have used GPU but it's actually running on the CPU

[uxhao-o]

I've checked to use the GPU, and I get a command with --device GPU in it, and then it seems to only use the GPU for the first few minutes, and then the cpu all the time after that.
By perform nvidia-smi within the docker container, I've found the GPU's memory stays close to 7G or so for the first few minutes when run program. Then, the GPU's memory usage quickly drops to the level that doesn't running program.

Another issue is that I processed T1w on the BIDS Validation [https://openneuro.org/datasets/ds004302/versions/1.0.1] dataset with GPU (actually cpu) specified, and it took me 4.4 hours to complete the processing of 3 cases (sub-01, sub-02, sub-03) were processed.