GPU folding bound by CPU, normal?

[danielocdh]

I wanted to fold with a couple machines that have 6 gpus each but they have dual core celeron cpus.
Usage as viewed on the Performance tab of task manager.
With 1 gpu running I get cpu usage around 60% and 1 gpu usage at around 50%
With 2 gpus running I get cpu usage at 100% and 2 gpus usage at around 50%
With 4 gpus running I get cpu usage at 100% and 4 gpus usage at around 25%
With 5 gpus running I get cpu usage at 100% and 5 gpus usage at around 20%

I removed the cpu slots on these machines because...well celeron cpus
I'm on fully updated windows 10, nvidia driver 441.20, Folding@home 7.5.1

Is this normal? and is there anything I could try to improve it?

[bruce]

Yes, that's semi-normal.

Windows expects a dedicated CPU for each GPU to move data through the PCIe bus. Linux does a better job of doing it with limited CPU resources.

[danielocdh]

So I made some tests with one system and ubuntu 18
I started using nvidia-smi to be able to better compare performance between windows and linux
The performance with my dual core celeron cpus was around 50% better on ubuntu, but it still isn't great(for 3+ gpus).
I'll finish the WUs and after that will be using 2 gpus on each machine for now.

[JimboPalmer]

Each GPU needs one CPU core. Celerons rarely have more than two cores