50 - 55c a safe overclock?

[hat]

It depends on how long... I ran Orthos blend priortiy 9 19 hours 30 minutes and some odd seconds

 

[imperialreign]

It depends on how long... I ran Orthos blend priortiy 9 19 hours 30 minutes and some odd seconds

that, too . . .

IDK, F@H requires the CPU to compute within a very, very small accuracy percentage. There is very little room for error. My typical OC test suite is running wPrime1024, SuperPI 32M, PCMark05 and ScienceMark2.0 all at the same time - which I've had all finish stable (if not extremelly slowed down ), but 5min of F@H has crashed the system at the same settings.

Which should be mentioned - the higher you OC, the less accurate a CPU's computations become.

 

[hat]

And the more voltage you supply, the more accurate they become...

 

[imperialreign]

And the more voltage you supply, the more accurate they become...

true . . . until you reach that architectural limit

 

[hat]

As long as it passes Orthos blend test for close to 20 hours I'm good.

 

[Graogrim]

It's not really an issue of more or less accurate--a CPU is a stochastic device. It is either correct as defined by specification or it is not. It doesn't make gradually worse approximations of the correct result--instead it's a statistical error rate. When the clock exceeds certain limits related to the physical and thermal properties of the circuit and the electrical current running through it, errors become more likely. And an increased voltage simply allows electrical signals to propagate better through the circuit pathways, sort of in the same way that increased pressure in a liquid or gaseous medium allows sound waves to better propagate.

 

[imperialreign]

I don't think we were trying to get that technical about it; but that's the heart of it right there