I can't find explanations of nsight compute metrices

Developer Tools Nsight Compute
1

Hello, I was profiling some kernels and encountered some metrics I can’t understand. The metrics are below

  1. derived_tempMetric0

  2. gcc__cache_requests_type_constant.sum

  3. dram__bytes_op_write.sum.pct_of_peak_sustained_elapsed

  4. gpu__compute_memory_access_throughput.avg.pct_of_peak_sustained_elapsed

  5. gcc__xbar2gcc_sectors.sum

  6. gpu__compute_memory_access_throughput.max.pct_of_peak_sustained_elapsed

  7. derived__memory_l2_theoretical_sectors_global_excessive

  8. derived__smsp__inst_executed_op_branch_pct

  9. gcc__xbar2gcc_sectors.sum

  10. gcc__average_cache_request_hit_rate.pct

  11. derived__smsp__sass_thread_inst_executed_op_ffma_pred_on_x2

  12. derived__memory_l1_conflicts_shared_nway

  13. derived__avg_thread_executed_true

I would appreciate if someone explains what the metrics above mean or notify me how to get access to the explanation of metrics above. I googled metrics above but couldn’t find any explanation.

Thank you in advance!

2

You can query metric descriptions using ncu --query-metrics or its sibling commands, or using the Metric Details tool window. You can also hover the metrics in the UI to see their description. derived__tempMetric0 is only used in calculating other metrics and doesn’t have a description.

Descriptions are also provided here.