pinning MPS client to unique core

Accelerated Computing CUDA CUDA Programming and Performance
1

In the official MPS guide, it is recommended to pin each client to a unique CPU core. It says this can be achieved by:

To accomplish this with MPS, launch the control daemon bound to a specific core, e.g.
taskset –c 0 nvidia-cuda-mps-control –d. The process affinity will be inherited by the
MPS server when it starts up.

However, wouldn’t taskset –c 0 nvidia-cuda-mps-control –d only pin the control daemon on core 0 but not the other clients on following cores ( ex. client0 to core 1, client1 to core2 … )?

2

That’s not what I read in the documentation. Please identify where that wording occurs in the latest MPS doc:

https://docs.nvidia.com/deploy/pdf/CUDA_Multi_Process_Service_Overview.pdf

3

oh, sorry for my misunderstanding. the exact quote is:

For setups where multiple GPUs are used with an MPS control daemon and server started per GPU, we recommend pinning each MPS server to a distinct core.

So it was recommending pinning the server and not he client.

Thanks.