Performance decreases after upgrading from windows 10 to windows 11

Hi, I’ve recently upgraded from windows 10 to windows 11. Then I found out the execution time of a function in my code which consists several gpu kernel function calls suddenly doubled. Anyone else experiences the same? Some advice or comment is welcomed. Thanks.

I also updated from CUDA 12.0 to 12.1 and newest driver . But it doesn’t make any change.

Got something seems to be the cause. When I turned off the Windows 11 feature “Hardware-accelerated GPU scheduling”, the execution time is same as was in Windows 10. It seems this feature would offload some amount of CPU work to GPU so that GPU is much busier while CPU has less workload. But this is apparently not friendly to a CUDA program in which all the intensive computing part is done on the GPU.

1 Like

How do you turn this feature off? I notice that my workload is twice as slow on Windows 11 as on Fedora - perhaps this is why.

When I googled for that text, I got this as the first hit.

1 Like

As Robert_Crovella pointed out, the way to run it off in Window 11 is: System->Display->Graphics->chang Default graphics settings, and then toggle off the “Hardware-accelerated GPU scheduling”.

I disabled GPU scheduling but it made not difference to performance.