Ideally, processor power consumption goes up with the square of the voltage and linearly vs frequency. Looking at your values
- 1206mV @1418MHz
- 1187mV @ 1405MHz
- 1168mV @ 1392MHz
- 1206mV @ 1527MHz
- 1187mV @ 1515MHz
- 1106mv @ 1455MHz
- 1093mV @ 1443MHz
we can find the ratio between the lowest overclock and the highest stock speed:
*(1443MHz/1418MHz)*(100%) = 83.6%
In other words, you've cut power use - and therefore heat production as well - by a sixth and gained (negligible) performance. But semiconductors also have an interesting quirk compared to, say, metals. When heated, their impedance decreases
and causes them to draw more
power (which heats them further, causing them to draw more power and heat...). That's one reason why AMD's Fury X can draw as little power as it does, and I think it's actually more efficient than the 980Ti or Titan X by a small amount. The temperature has a hard limit of 65°C, much cooler than about any other GPU on the market, which reduces wasted power.
Undervolting is a tried and true tactic for lowering a system's power. Stock voltages, I believe, are there so every single piece of silicon shipped out is guaranteed to hit those frequency targets. Even if you get an average processor, undervolting it is feasible to some degree. AMD's CPUs are really bad about it. FX 8 cores can hit like 1.1-1.2V with a bit of effort, down from the stock 1.3-1.4V, which with no other changes cuts power by 15% minimum.