DCGM Memory used


nvidia-smi offers a memory.used metric that outputs “Total memory allocated by active contexts”. We are interested in tracking this but for a partition of a MIG (i.e., A100 40GB with a 1g.5gb profile). The supported metrics by DCGM include Memory BW Utilization, but it is different.

Is there a metric (or possibility) of tracking used memory in DCGM for a partition?


Bumping this.

It would be great if someone confirms whether memory.used exists in DCGM with the same definition used in nvidia-smi, and not in the sense of BW Utilization.