Basically yeah, you could increase voltage all you wanted but then you would have to decrease the frequency or somehow increase the spacing between Vcc and Vss or somehow decrease the transition time from Vcc to Vss and the opposite without having operational errors.
Sounds like you end up creating an infinite loop. By pushing your voltage too high you just create a catch-22 situation you can't get out of.So to make it more clear because I realize I omitted a diagram to show what happens when you overclock beyond a stable threshold. So this happens:
The voltage transition from low to high does not complete before the end of the cycle, and it doesn't reach to the recognizable ON (Vcc) threshold, and there you have your errors that start showing up as artifacts, freezes, etc. (or wrong results if you run something like Intel Burn Test).
That is why you increase the voltage. To "push" the transition to a higher level faster, but that will increase temperatures and the increased temperature will increase resistance, making the transition harder, thus requiring more voltage, which will in turn produce more heat, which will in turn produce even higher resistance increasing the transition curve further, and so on and so forth...
Well, with extremely good cooling you can avoid most of it by keeping resistance in a very low level. This is why you can see liquid nitrogen cooled systems reach extreme stable overclocks. Any by keeping the temperature low, you don't need to increase the voltage too much since you still have better (faster) transition from low to high states.
Just stick a 290x in it, that's the fastest way to set anything on fire.
