I am looking for equations for calculating utilization of the shared memory, l1, l2, and hbm of a H100. I generated a ncu-rep file with --set detailed to see the memory table. I confirmed this table exists by using a ncu-rep file I found on this forum taken by an rtx3060 ti. I was able to mouse over the Peak % columns and get an equation.
However I was worried these equations might have changed for the H100, so I generated the file on H100s that I am profiling. This time when I loaded the ncu-rep file the table was missing.
The memory tables are not part of the “detailed” set, you need to collect --set full. See the file MemoryWorkloadAnalysis_Tables.section in the ncu sections directory for reference. You can also use ncu --list-sets or ncu --list-sections, or the respective UI tool window to get the information.