Effect of driving display on PyTorch performance

I have a computer with both a NVIDIA Geforce RTX 3070 and an Intel UHD Graphics 630. The Intel chip came disabled, with the manufacturer suggesting that the NVIDIA drive the monitors.
If I use the NVIDIA GPU to drive my displays (Windows desktop, IDE, web browser, etc), is that taking any appreciable amount of resources away from the NVIDIA being able to do PyTorch/CUDA work? Would it be beneficial to have the Intel GPU run the desktop so that the NVIDIA can be dedicated to PyTorch?
One downside of using the Intel GPU is that it consumes system memory, and uses electricity power. The power means heat, so the CPU can’t run as hard.


A disadvantage of using your 3070 for the display is that you would be using GPU memory, which is usually more limited than the host RAM.
E.g. in my plain Ubuntu20.04 setup my GPU is using ~1.2GB for the display output (Xorg, gnome-shell, firefox etc. are reported by nvidia-smi).

1 Like

Thanks, that helps. I didn’t know about nvidia-smi. I also see that the NVIDIA’s memory is being used.