Nsight Compute failed to profile GPU data with [Error] float division by zero

when launch the profile in Nsight Compute, something wrong occurs in the Error Report as follows.

Issue Slot Untilization: [Error] float division by zero
Error path: NVIDIA Nsight compute/2021.1.1/Sections/IssueSlotUntilization.py:52

Then I go to the 52nd line of the code, the denominator is IssueActive, the definition of which is at the 44th line:
issueActive = action.metric_by_name(“smsp__issue_active.avg.per_cycle_active”).as_double()

So it means that the number of the operations is zero?
(I think it is impossible…… and I would appreciate information about this error!!!)

As you found, the root cause is that the value of the smsp__issue_active.avg.per_cycle_active metric is 0 in your collection, which then manifests to this div-by-0 error in the python rule. It is not expected that this metric is 0, but it’s also not immediately clear from your description why that would be the case.

My recommendation would be to simply move to the latest Nsight Compute version, 2022.1.1, which has many bug fixes and new features compared to the one you are currently using. Newer tool versions are backwards-compatible with older driver and CUDA toolkit versions.

Thank you so much. I update it to the latest version and obtain the profile report successfully.