You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Can you extend the metric argument in nsight.analyze.kernel to support collecting all metrics from a set? Manually picking the metrics for each section in the set is cumbersome. I understand that this slows the benchmarking down and results in bloated output csv files, but would like to have this as an option when I need detailed ncu reports.
@arjunmenonv would you want to collect the metrics from a set for different kernel configurations and multiple runs for each configuration as supported for a single metric?
Or this would be used only for a single configuration and single run?
Hi @nvssatoor, sorry I missed your reply.
Ideally I'd like to collect all the metrics for a single kernel configuration with a single benchmark run, like we do with the Nsight-compute cli tool. Support for different kernel configurations is nice to have, but I am ok writing some python wrappers / bash scripts to do this at my end.
Activity
nvssatoor commentedon Dec 18, 2025
@arjunmenonv would you want to collect the metrics from a set for different kernel configurations and multiple runs for each configuration as supported for a single metric?
Or this would be used only for a single configuration and single run?
arjunmenonv commentedon Mar 17, 2026
Hi @nvssatoor, sorry I missed your reply.
Ideally I'd like to collect all the metrics for a single kernel configuration with a single benchmark run, like we do with the Nsight-compute cli tool. Support for different kernel configurations is nice to have, but I am ok writing some python wrappers / bash scripts to do this at my end.
nvssatoor commentedon May 27, 2026
@arjunmenonv we are evaluating enhancing nsight python to support more metric collection options. This should also cover your request.
Thanks