Hi, I see multiple “cuda-EvtHandlr” CPU threads when running my application, is it possible to have some information about what they actually do, why they’re more than one and if it’s possible/suggest to move them to an isolated core to decrease latencies?
I may have been not very clear. I don’t know of any published information that explains how CUDA uses the host threads it may spin up. I’m not able to offer any confirmation of anything, other than that they do exist (it is expected behavior). You can always request an update to CUDA documentation by filing a bug. I don’t have any further information to share about it.