Hello.
I have encountered a problem with MPS.
If there are 2 client processes sharing one GPU through MPS and both clients have 90% of CUDA_MPS_ACTIVE_THREAD_PERCENTAGE, there must be interference between the client processes.(assuming both client process is conducting intense computation). What I want to ask you is that if i can control the level of interference.
For example, If client process1 and client process2 have CUDA_MPS_ACTIVE_PERCENTAGE 90%, 30% each,
I want process 2 to obtain full mps percentage 30% when proces2 is conducting intense execution but process 1 to have 90% mps percentage when process 2 is not conducting intense computation. Like this, can is there any way to control the level of interference between client processes on MPS?
Sorry for my short english and thank you in advance!