Thank you for your reply.
I just observed something very strange and I would like to ask you:
I run the opencl process ./program and with nvidia-smi I observed 6% GPU utilization.
When I run ./program & ./program I observe that the full GPU utilization is 30%.
I do not really understand why.
Theoritically, as concurrent processes are not supported, there should be a slicing scheduling between the 2 processes right? That means that the used SMs should be 6% right? (the same percentage of the single process as only one process is running on the GPU at any time).
Additionally, if somehow there was spatial scheduling (both processes fit on the GPU as there are available resources) the final GPU utilization should not be 6%+6%=12%?
Why it is 30%?
Could you please explain me what is happening?