For a given NVPModel,
we can observe the frequency of a particular unit (be it CPU, GPU or Memory) changes based on demand/utilization. When the utilization falls, the frequency does not go down immediately rather it waits for some time and scales down.
So one can assume, the dvfs is called periodically to adjust the frequency of each unit.
I have two questions:
how frequently the dvfs functionality is activated to check on the utilization?
And once changed, how long does it wait to scale down when the utilization drops?