I wanna profile GPU SM active and SM occupancy to view acutal load of GPU. As we know, nsight or nvprofil can get this metrics, but this tools are bind with CUDA application. How can I get this metrics from device aspect like dcgm or nvml api? NVML API had already support SM active and SM occupancy on Hopper or newer devices(NVML API Reference Guide :: GPU Deployment and Management Documentation). However, Ampere architecture is not support. what should I do to get real GPU util for my nvidia A10?
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| How to monitor SM utilization and SM occupancy? | 7 | 9590 | January 12, 2024 | |
| Measure SM utilization per process | 1 | 1143 | January 11, 2024 | |
| What is the lowest level of GPU application that can be monitored on a GPU/GPUs? is it PID or threadId? | 4 | 274 | July 5, 2024 | |
| I want to get only GPU Utilization | 1 | 472 | August 12, 2021 | |
| Query if GPU is occupied | 1 | 900 | January 22, 2013 | |
| nvidia-smi power draw unspecified | 0 | 1505 | January 5, 2018 | |
| Watch Resource Usage of an SM in Real Time | 1 | 648 | April 12, 2023 | |
| How to monitor GPU performance (usage, memory and Tx/Rx) in realtime? | 2 | 3956 | July 29, 2019 | |
| How to get gpu Utilization with nvmlDeviceGetUtilizationRates() | 2 | 1675 | July 18, 2019 | |
| How to know that how many threads were located on a device at a certain moment or during a period? | 0 | 521 | April 14, 2015 |