have to use offset voltage and not manual to get the Vcore to lower.
anyways, I've had now two sandy bridges with two different motherboards. (gigabyte UD4 and Asus Deluxe). here's my personal experiences.
my first an "A" batch, OC'd like poop. needed internal PLL for anything over 4.5GHz, and needed 1.38V for 4.6GHz. above that, it needed insane voltages for 4.7GHz/4.8GHz like around 1.46v, but I could never get it stable at 5.0GHz even with 1.48v-1.51v. it ran 4.8GHz for a week at 1.44-1.64v before degrading to the point where it wouldn't even run stock clocks with my 2133MHz memory anymore. (had to run 1866 or lower)
my second, a "C" batch is pretty decent but not amazing. needs internal PLL for anything over 4.7GHz. does 4.5GHz 1.3v, 4.7GHz 1.376v, 4.8GHz 1.4v, 4.9GHz 1.43v but 5.0GHz needs something over 1.45V which I'm not comfortable using so I didn't even test. (I'm guessing somewhere around 1.476-1.5v for 5.0). so my dream for that magical 5.0 number was just a pipe dream
so, consider yourself VERY lucky if you have a chip that will do 5.0GHz with 1.41v. Asus says in their testing of 100+ CPU's only about 2% of them were able to achieve 5.0 and higher and typically in the mid 1.45v range.
and a note about voltage, I'd consider anything 1.425v or lower to be fairly safe as long as your temps are <75c, preferably 70c or lower. but if you're constantly beating it up with torture tests or BOINC then I'd prefer 1.4v or lower. what causes electromigration is the combination of heat and
current. (i.e. higher voltage = more current. and higher MHz = more current too)
for stability testing I've found that medium size FFT's and Prime "blend" seem to find instabilities the most. see, the decode engine in the chip seems to be the weak link and this hits it harder. (I typically do 1024KB FFT's with 4096MB) meanwhile, IBT/LinX just hit the FPU very hard but prime still seems to hit the weak points better.Edited by s74r1 - 3/18/11 at 2:59pm