Higher voltage requirement for lower clock frequency

2

0

Is there any reason for a CPU to require a higher Vcore to achieve stability when running at a certain speed, than the Vcore that the same chip requires for stability when running at a higher speed?

Specifically, I am (moderately) overclocking my Ivy Bridge i7 3770K. I am not touching the BCLK, leaving that at the default of Auto (100.00 Mhz), turning just the multiplier.
When I have the multiplier set at x44, the system is perfectly stable (judged by stressing with Prime95 for a few hours), with the Vcore turned down to 1.200v.
If I dial the multiplier back down to x43, the system is no longer stable, and crashes after less than half an hour of Prime95. In order to be stable, I need to crank the Vcore back up to 1.220v.
(This happens regardless of what the VCCPLL is set to, this happened when it was on Auto (1.800v), manually set to 1.800v, and also when it was pushed down to 1.505v - so this seems irrelevant at this point).

Admittedly, this is not a huge difference, nor is it a very big deal - 1.220v is still agreeably low.
However, I am bothered because to the best of my knowledge, higher frequency should require higher voltage.

Why would my chip need a higher Vcore at x43 than it does at x44?

(If it matters, my motherboard is a GA-Z77-D3H.)

AviD

Posted 2013-04-16T16:22:30.887

Reputation: 481

Answers

3

The short answer is this: Signal integrity is hard, and the rule of thumb higher frequency should require higher voltage is not the whole picture.

There are lots of factors that go into high-frequency signal integrity and without looking at the actual failures at an electrical level, it can be very difficult to determine what the actual cause is. Off the top of my head, here's a couple of possible reasons:

  • Temperature. When pushing the boundaries of signal frequency, the temperature has a very real effect. What was the temperature of the CPU when you were doing your tests?
  • Crosstalk. I don't know what other signal frequencies there are on your CPU, but 44x is nice and round which may increase the crosstalk due to harmonics. Look at "Measuring Crosstalk in the Frequency Domain" on page 4 of this PDF on Crosstalk Measurement.

There are of course other possible reasons why your CPU requires a higher voltage at a lower frequency, and the possible reasons that I gave may be wrong. Once again, signal integrity is hard.

Here's a couple of quick references:
Wikipedia: Signal integrity
The Basics of Signal Integrity
Optimized Signal Integrity

Tim

Posted 2013-04-16T16:22:30.887

Reputation: 1 375

Thanks! The temperatures were higher on x44 (as expected), but still within reason (<80C). On x43 they were quite a bit lower. That crosstalk pdf was quite a bit over my head - I agree that it is hard - but if crosstalk was increased, wouldnt that be the more unstable setting? – AviD – 2013-04-16T20:56:05.743

I guess the bottom line is as you say - signal integrity is hard, and there are probably other factors in play here. Still, I'd like to understand a bit - maybe those other links will help. – AviD – 2013-04-16T20:56:46.057

1

I wish I could explain it better, but I only know enough about the subject to know that it's very difficult. Perhaps you'd find a better answer on http://electronics.stackexchange.com

– Tim – 2013-04-16T23:12:39.677