Specifying multiple GPUs (e.g., -g 0,1,2,3) still uses only a single GPU

[jair-pereira]

I ran the following command on the test dataset:

deepmasc -f examples/Class3D/job016/run_it025_class*.mrc -g 0,1,2,3 -o output

However, I believe the script is still using only one GPU:

• The running time is essentially the same as when using -g 0.
• nvidia-smi shows activity only on a single GPU during execution.

It appears that providing multiple GPU IDs with -g is not enabling multi-GPU usage as expected.

This issue does not happen when running through the Relion External tab.

[AntiMatter568]

I'm not able to reproduce this situation, have you tried setting the environment variable using export CUDA_DEVICE_ORDER=PCI_BUS_ID?