Importance of number of CPU cores

Hi, im Sukyung

I want to ask about the relationship between GPU and CPU. I’m going to build a new computer for GPU-based calculations. I wonder if the number of cores is important when calculating GPUs. Now, I am using 4*RTX 3090 ti GPU and doing ILES calculation through Backend cuda.

Does each gpu use one core unconditionally?The processor number changes, but four always seem to be 100%.

I have seen an increase in the overall cpu share periodically

Is this related to [soln-plugin-tavg] or [soln-plugin-fluidforce?] … (soln-plugin]?
If it is related to this, does the more CPU cores, the less calculation time when i use backend-cuda? Does the low cpu core take a long time or affect the overall calculation time?

Generally, when running on a GPU based platform, PyFR needs one core per GPU. Some plugins are able to indirectly make use of more cores, although the benefit here is not typically substantial.

Regards, Freddie.

When one GPU uses one core of CPU, does the performance of the CPU also greatly affect the simulation time?

So long as each GPU has a reasonable amount of work to do (above say 10,000 elements) then CPU performance is not important. If the GPUs are less loaded it can have a small impact.

Regards, Freddie.