Thanks for the answer.
But is it not contradictory that they will run at same time but serialized? I mean if they are serialized then how would they run at same time?
Is there some gap in my understanding?
Also as per my study it seems even with MPS, it allows for different kernels from the same applications to run concurrently if the kernels are having the same CUDA context. But if they belong to different applications then the context can not be same and the tasks are effectively serialized. Isn’t it?
Regards
Deepak