does not support properties of utilization metrics for MIG devices, according to Is there another way to calculate the usage rate of gpu?
Under MIG mode, only DCGM is capable of showing GPU utilization. DCGM can even show per-GPU-instance and per-GPU-Compute-Instance utilization.
Once you install the package try:
sudo nv-hostengine
dcgmi dmon -e 1001
Then “sudo nv-hostengine -t” to stop nv-hostengine.
1 Like
Thank you for your answer.
The result screen is as follows, but is the GRACT item in use correct?
Additionally, I would like to connect with other programming languages such as Python through DCGM, is there any example or documentation to refer to?
Yes. Graphics Activity (1001) is your baseline GPU utilization.
There are more GPU activity metric descriptions here:
There are C and python samples installed with the dcgm package.
Nice! Thank you very much.
Additionally, is it possible to collect GPU usage by multi-instance GPU profile?
I’m still looking for that document, but I don’t think so.