Folding Forum

So, on the chrome client, when I drag the bar to max, it uses 100% CPU. This is fine. When I drag the bar to max on the desktop client, it uses 44% of my CPU, and 6% of my GPU. Help?

This would be easier with a log of what the program thought it was doing and what it thought it was running on. Usually the first 100 lines of the log tell this.

viewtopic.php?f=24&t=26036

Tuning your client for performance has to start with knowing what it is doing now.

SepticGengar wrote:So, on the chrome client, when I drag the bar to max, it uses 100% CPU. This is fine. When I drag the bar to max on the desktop client, it uses 44% of my CPU, and 6% of my GPU. Help?

Folding processing done on you CPU are accounted separately from processing that's done on your GPU. Your figures make sense, but you have to interpret them correctly.

The 100% of your CPU figure means that the Chrome client is using all available CPU resources (and none of your GPU which is probably idle).

When you enable GPU folding, the number that is important is what percentage of the GPU is being used. That information does not show in your figures. Take a look at what the Sensors tab in GPU-Z reports (or other similar programs) or one of the GPU diagnostic programs supplied by people who produce GPUs. Each GPU requires the services of one CPU thread to transmit/receive data between main RAM on the CPU and the VRAM on the GPU. Since all the "heavy" calculations are being done on the GPU, the CPU thread only needs to move data plus an occasional validation calculation so that CPU thread isn't very busy.

The GPU does those heavy compute processes MUCH, MUCH faster than your CPU can and it uses very little of the CPU to do it.

If your CPU has J threads and you have K GPUs, you can use as many as (J-K) CPU threads to do the same kind of CPU processing that the Chrome client does but a single GPU will do a lot more science and earn a much higher PPD.