When I tried to use ncu to profile the python script, a lot of “NAN” appeared, but the same command before was not “nan”. I was very confused. Is it because of the python script? Or is it because of the cuda driver?
Can you clarify what you mean by “but the same command before was not “nan””? Was it a different version of the tool? Or profiling a different workload that didn’t use python? Is this a repeatable case where you will always get “nan”? Can you share what version of the driver and Nsight Compute you are using?