Hi Cuda/MPS team,
I have a 2 gpus on my system.
I have enabled MPS server setting CUDA_VISIBLE_DEVICES=0 which means only 1 gpu is used by MPS.
I want to run a few related applications on GPU 0 which is controlled by MPS.
And I want to run another unrelated application on the other GPU i.e. gpu 1 which is not controlled by MPS. However, i see that the application gets intercepted by MPS server thus running on GPU0.
Is there a way to tell a cuda application to not communicate with MPS server?