Help me overclock/cool Ivy Bridge

[Yuriman]

Err, those really wouldn't be additive. If anything, if the Intel tool loads a profile after Windows starts, your offset is .035V (not mV).

Tested and they are in fact additive. Changing offset voltage from -0.025v to Auto in bios resulting in a change in load voltage as reported by CPU-Z from 1.280 to 1.304v. This assumes CPU-Z isn't reading incorrectly, of course.

EDIT: I imagine running like this has some advantages. The offset in bios should apply to all clocks, while the one set in Windows applies only to boost voltage. Effectively I am running with a small undervolt at non-boost clocks and a slight overvolt at 4.5.

 

[coffeejunkee]

Also, Linpack + Prime seem to be quite insufficient for stability testing. I left Linpack running overnight at 4.5ghz/1.262v and it was running when I woke up, and Prime doesn't crash either, but I get force-closes and crashes in games within 5 minutes unless I add another 0.010v.

This is exactly my experience. For instance I could do 20 runs Linx/hours of Prime at 4GHz using -0.070 offset. But I had crashes and lockups playing Forged Alliance and had to bump to -0.055 to be stable.

FA seems like best test since this game is very heavy on the cpu. 3DMark Vantage fex would run fine at the same vcore needed for Linx/Prime stability. But again crashes in FA.

Same at higher frequencies until 4.4GHz where it remains unstable even with liberal extra volts. Seems it's the c-states which cause trouble at that frequency and higher.

 

[sm625]

I've long suspected that c-state changes might catch problems that a constant load might miss. If changing c-states causes stability problems, then why hasnt anyone modded a prime95 type program so that it pauses every second to let the c-states and p-states change?

Also, when I installed a 3770, it was clear that the IHS on that chip was very tall. I dont remember any other chip with such a high vertical height, except perhaps a windsor era athlon. And boy did those things run hot too.