The meaning of total wall clock time and other metrics in dlprofviewer

Developer Tools CUPTI – CUDA Profiler Tools Interface
1

AFAIK dlprofviewer calculates the total wall clock time and other useful metrics using nsight systems nsys_profile.sqlite. Also, It will be very helpful if there is documentation on how the different metrics shown in dlprof are calculated from CUPTI tables.
Update
I got a match of the GPU utilization metric shown in dlprof and the sum of entry-start for all kernels in CUPTI_ACTIVITY_KIND_KERNEL. I got the match of the Memory utilization metric shown in dlprof and the sum of entry-start for all entries in CUPTI_ACTIVITY_KIND_MEMCPY.

But there is no entry in the tables for the CPU time. The question then is how is dlprof calculating the CPU time for the ops? Thanks.

2

@puneethnaik please be advised that DLProf has been sunset and unsupported for many months now. We recommend that you directly use nsight systems or a native framework profiler.

3

Got it. Thanks for the clarification. I will use nvidia nsight systems from now on…

4

Sad news -_-