I am trying to get GPU utilization of all processes, currently running on GPUs. Using NVML C APIs, I can get all the process PIDs even if there is any process which has started just now. It seems I cannot get the same with DCGM, does DCGM C APIs always need to “provide” PID externally? Can’t it discover a new process PID on its own?
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Unable to retrieve running processes in DCGM | 0 | 950 | April 25, 2019 | |
| DCGM Not reporting running processes | 1 | 604 | April 25, 2019 | |
| DCGM exporter does not export mps process id's where as it shows in nvidia-smi | 0 | 131 | April 9, 2025 | |
| Provide solution for "GPU MEM used by PID but no GPU LOAD" | 2 | 2285 | January 10, 2023 | |
| Get GPU Power Usage per PID in python on linux | 1 | 1860 | October 31, 2021 | |
| `nvmlDeviceGetComputeRunningProcesses_v3()` inside container reporting invalid memory usage number | 0 | 429 | May 30, 2024 | |
| How nvmlReturmlDeviceGetAccountingStats figure out gpuUtilization if more than one processes run on one gpu? | 2 | 812 | November 10, 2023 | |
| How can I get the PID onGPU of the current program in code？ | 0 | 357 | May 19, 2021 | |
| Getting wrong values for 5 fields retrieved using DCGM Process Statistics C APIs | 0 | 149 | July 5, 2024 | |
| How to monitor per process's utilization on GPU under mps | 3 | 828 | November 5, 2024 |