Quote Originally Posted by ugotd8 View Post
Well, I've only understood half of what you have said so far, but thank you very much for shedding some light on this. I've noticed the peculiarities of stress testing on this new chip and I'm frankly a little lost as to when/how to call an overclock stable.

I see people on other forums hammering their chips with voltage trying to get Prime95 to pass. I on the other hand am holding off until this is sorted, it just doesn't make sense to me that on the 1366 platform the linpack test was the killer (highest load) and now on the 1155 platform that has miraculously switched around and now Prime95 is the killer.
Feel free to ask about what you don't understand.

Different architectures have different failing mechanisms; on Nehalem the execution units (add, multiply, load/store) were the bottlenecks, but Sandy Bridge it seems the instruction decode is the bottleneck.

Linpack is designed to be more execution stress whereas Prime95 could possibly require more decode stress (especially if you choose blend instead of small fft).

I cannot verify this (still waiting for my Sandy Bridge), but if more people have problems with Prime95, that might be the goto test until more evidence comes in.