CUDA Visual Profiler with multiple GPUs

Hi all!

I use two GPUs concurrently by multithreading. The program runs fine.

But I’m somewhat confused by the result of the visual profiler. Could anyone tell me, which one of the two GPUs are profiled? or both?

Doesn’t it say which device you’re running on in the Sessions tab?

e.g. Session1 ->Device_0


Ahh, yes!
Session Properties -> CUDA Device(s) -> here lists my two GPUs.
So this is calculated by the latest end minus earliest start?
Oh, it does’t seem to be much shorter than using only one GPU… this might be anogher problem?