Multiprocessor activity represented by guided analysis

Accelerated Computing CUDA CUDA Programming and Performance
1

In the nvprof, I see this chart for multiprocessor activity.
https://pasteboard.co/ITUNcQp.png

However, the nvprof data shows
1459 sm_efficiency (80 instances) Multiprocessor Activity - - [ 58.13% 0.00% 0.00% 0.00% 0.00% 0.00% 16.96% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 37.67% 0.00% 0.00% 0.00% 0.00% 0.00% 13.95% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 33.79% 0.00% 0.00% 0.00% 0.00% 0.00% 10.45% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 24.64% 0.00% 0.00% 0.00% 1.10% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 27.25% 0.00% 0.00% 0.00% 0.00% 0.00% 5.53% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 19.34% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% ]

The chart shows only one SM. However, there are some other SMs that have some values. For example, 33 and 24 percents should be shown.

These are verified for one kernel name. Are there some reasons for that, or there are some mistakes by me?

2

The chart is showing the value for a single grid launch. I believe the nvprof is showing you the aggregation across multiple launches of the kernel.

3

The chart is showing the value for a since grid launch.

Excuse me. I didn’t understand that…
Do you mean the first launch only?

4

Fixed to say “single grid launch”. In NVVP the chart is displayed for the selected launch. The command line tends to roll-up stats for all grid launches with the same kernel name (independent of launch dimensions or other criteria).