Is CPU utilisation a good proxy of CPU power consumption?

[ohuu]

Is CPU utilisation a good proxy of CPU power consumption let alone system power consumption?

Until now Cardamon has been using CPU utilisation as a proxy for power consumption. We take the TDP of the processor and assume that 100% utilisation would consume that amount of power. Then we take the utilisation caused by the processes we observe and work calculate the power consumption of that process accordingly.

However, this Quora discussion and other resources suggest that the correlation is pretty weak.

If this is true then CPU utilisation may lead to some seriously confusing results.

Might it be better to profile a system regularly by observing power consumption of the CPU directly under conditions where the process you want to measure isn't running. Then take readings with it running and use statistics to work out how much of the power fluctuation was caused by the process?