MPS (Multi-Process Service) in two GPUs

Hello, I have a workstation with two T4.
I run the same code with multiple processing framework in the two cards in MPS model.

I run “nvidia-cuda-mps-control -d” in the two cards respectively, and then start the code.

The code are run with a particular GPUs using the CUDA_VISIBLE_DEVICES variable. The processes running in a card work parally but the others in another card do not.

Here is the output of nvidia-smi. I find the type of some processes is M+C but others is C. Is there anything wrong? Thanks in advance.