Hi,
I use a local Katago opencl executable with Katrain to analyze games. However, when I run the engine, the GPU usage is pretty low, rarely over 10%. I expected a high GPU usage, is that just a fallacy?
I tried to adjust some settings in the config, but that doesnt really change anything.
Is that just nomal?
Thanks
If you’re on windows, you should entirely ignore window’s built in GPU usage measurement (and if you really care, search around online for what ML-related GPU benchmark people actually use).
Because as I understand from what other users have reported, window’s normal GPU measurement mostly looks at some things relating to actual graphics rendering like a video game would use and ignores lower-level matrix operations even if those things are exercising the GPU heavily… and of course a neural net for Go is doing raw matrix math on giant tensors of neural net activations, rather than doing fancy 3D graphics and shaders.
Nice. Thank you very much! 
I will test it
Yes, it is exactly how you said it.
Not that I ever doubted you, I just wanted to test it regardless. ^^