Why some metric sets in ncu is not enable?


Identifier Sections Enabled Estimated Metrics


default LaunchStats, Occupancy, SpeedOfLight yes 36
detailed ComputeWorkloadAnalysis, InstructionStats, LaunchStats, MemoryWorkloadAnaly no 173
sis, Nvlink, Occupancy, SchedulerStats, SourceCounters, SpeedOfLight, Speed
OfLight_RooflineChart, WarpStateStats
full ComputeWorkloadAnalysis, InstructionStats, LaunchStats, MemoryWorkloadAnaly no 178
sis, MemoryWorkloadAnalysis_Chart, MemoryWorkloadAnalysis_Tables, Nvlink, O
ccupancy, SchedulerStats, SourceCounters, SpeedOfLight, SpeedOfLight_Roofli
neChart, WarpStateStats
source SourceCounters no 56

When i use “ncu --list-sec” to see the metric set in my nsight compute 2020.3.0. with CUDA 11.2, it seems that some metric sets are not enable, what is the reason?

In fact, I can not see the not enable metric sets (for example, MemoryWorkloadAnalysis) in ncu or ncu-ui even through change user to root or run by sudo.

I run it in user (no root but sudo) on Ubuntu 20 with 4 RTX 3090 cards. The driver version is “NVIDIA-SMI 470.74 Driver Version: 470.74 CUDA Version: 11.4”, which is printed by “nvidia-smi”.

The “ncu --list-sec” shows whether a section is enabled in the default set or not.
If “–set” option is not specified, the default set is collected. The full set of sections can be collected with “–set full”. Or you can collect data for a specific section using the “–section” option.

I know this option’s meanings. But my question is that wether some metric sections are not enable in my device. When I use “ncu --list-sections”, the print message shows that some sections are not enable in the Enable column. In fact, I can not see this metric sections (for example Memory Workload Analysis) in the report even through i use “-set full” option. What is the reason and how can I solve this problem? Thanks.

The fllowing is the print of “ncu --list-sections”:

Identifier Display Name Enabled Filename

ComputeWorkloadAnalysis Compute Workload Analysis no …21.2.0/Sections/ComputeWorkloadAnalysis.section
InstructionStats Instruction Statistics no …2021.2.0/Sections/InstructionStatistics.section
LaunchStats Launch Statistics yes …pute/2021.2.0/Sections/LaunchStatistics.section
MemoryWorkloadAnalysis Memory Workload Analysis no …021.2.0/Sections/MemoryWorkloadAnalysis.section
MemoryWorkloadAnalysis_Chart Memory Workload Analysis Chart no …0/Sections/MemoryWorkloadAnalysis_Chart.section
MemoryWorkloadAnalysis_Deprecated (Deprecated) Memory Workload Analysis no …tions/MemoryWorkloadAnalysis_Deprecated.section
MemoryWorkloadAnalysis_Tables Memory Workload Analysis Tables no …/Sections/MemoryWorkloadAnalysis_Tables.section
Nvlink NVLink no …Nsight Compute/2021.2.0/Sections/Nvlink.section
Nvlink_Tables NVLink Tables no …Compute/2021.2.0/Sections/Nvlink_Tables.section
Nvlink_Topology NVLink Topology no …mpute/2021.2.0/Sections/Nvlink_Topology.section
Occupancy Occupancy yes …ght Compute/2021.2.0/Sections/Occupancy.section
SchedulerStats Scheduler Statistics no …e/2021.2.0/Sections/SchedulerStatistics.section
SourceCounters Source Counters no …ompute/2021.2.0/Sections/SourceCounters.section
SpeedOfLight GPU Speed Of Light Throughput yes … Compute/2021.2.0/Sections/SpeedOfLight.section
SpeedOfLight_RooflineChart GPU Speed Of Light Roofline Chart no …2.0/Sections/SpeedOfLight_RooflineChart.section
WarpStateStats Warp State Statistics no …e/2021.2.0/Sections/WarpStateStatistics.section

Can you please provide more details? Do you see any errors with the “–set full” option? Hope you are not using “–list-sections” along with “–set full”? Please try and share your complete command line & the ncu output.

You can try if “–section MemoryWorkloadAnalysis” works.