Multiple tensorflow sessions running on the same GPU in parallel (with/without MPS)

Hi, I have this curiosity which I hope you will help resolve :)

My use case is where different threads in the same process run different tensorflow jobs on the GPU. These tensorflow jobs have their own session and they might try to run on the GPU at the same time.

I think by defaullt these tensorflow executions will time-share (meaning get preempted and switch between each other). Note that the resource usage of these jobs is pretty small (around 30% of memory and do not fully utilize the gpu compute).

Now I was wondering what happens when I enable MPS? Do these tensorflow jobs run concurrently (without time-sharing) even if they originate from the same process?

Thank you.