OOPS! Sorry for that 1.162v I mentioned - that's the voltage while in BIOS. The 1600 to 4600Mhz OC runs between V1.000v to 1.334v with a - 0.005 offset while in Windows.
You also require a descent OC's MB. I'm running an ASUS GA-P8Z68V-Pro Gen3 which is a very friendly OC'r plus I'm water cooling with temps at idle of 34 maxing 47C under Prime95 for that clocking.
Temps for the constant 5000Mhz clocking with 1.462 to 1.521v's and + 0.115v offset is 42C at idle to 67C running Prime95.
Personally I don't like going over 1.5v on a SB processor as it could damage the CPU's memory controller but I've heard lots do it without issues.
That i7 2600k would love a Corsair H110 and easily run at 4600 to 4800 Mhz's and again a descent OC'g MB with a good PSU.
I've got the first release of that board, and I'm not entirely sure of the difference. I thought I'd read where the Gen3 board was "ready for Ivy Bridge." Later, I thought I saw BIOS upgrades for my Pro board which included " . . to run 22nm" or "make compatible with new processors."
I have a friend -- retired physicist; we'd both sat in the same Thermo, Nuclear Physics and 4A-thru-4D freshman introduction classes. He still tells me the limit for a transistor is 1.5V. How this applies in some ballpark with the Intel processors, he couldn't likely explain, and I wouldn't know.
So -- no objection to your high benchmark clock. I just think it's marginally useful to push the processor over either a mfgr spec or a general consensus for "safe" voltage. Intel hadn't included that spec since Nehalem, if I'm not mistaken. It was ~1.38V. If the previous Wolfies and Yorkies were 45nm, the Nehalem has a lithography like SB or 32nm. There's some archive web article saying that Intel scrapped 45nm for Nehalem, and only the 775's mentioned, the Lynnfield and Clarksfield are 45nm. Other socket 1366 Nehalem cores were 32nm, like Sandy Bridge.
1.38V was also the safe range upper-bound spec'd for Wolfdale and Yorkfield.
So it may be that 1.38V is a reasonable safe limit for Sandy Bridge even though it's not a published spec, and this had been assumed in some reliable web overclocking guides. Even so, some subset of SB enthusiasts came to a consensus it should be 1.35. But any such official specs are a judgment about probability frequency distributions, RMA cost-accounting and other factors. In many of these respects, it is simply a line bisecting some probability density function of failure within the 3-year warranty, and the thinking behind it is to make the chances of warranty replacement unlikely enough to be insignificant or near-zero.
In my case -- 30 months after I built this rig -- I'm trying to eliminate the very most infrequent idle EIST instability. But I use the same techniques to hone in on the optimum voltages (among all those that are stable or near-stable) to get the maximum GFLOPS with the smallest variation. So I can take sample runs of 15 iterations or so, save the LinX logs, load them into the spreadsheet and put in a couple cells for each list using "AVG" and "STDDEV." Or, more crudely, I can look at the absolute range (Max less Min), and I can even throw away outliers that are explainable. For instance, I may observe that Media Center is doing a background update of program listings during a particular run which shows 126 GFLOPS when all the rest are between 127 and 129.
BAsed on these things, I might adjust voltage 4mv up or down, or 5mv up or down in "offset" and "Turbo Voltage" for the sweet spot. These will always be higher than voltages found just adequately stable and bumped up a notch for "safe measure."
So I might go to 4.8Ghz if I can further lower the sweet spot for 4.7, and if I venture into that higher failure distribution, I could only guess whether CPU life is shortened two, three, four years and so on. If I could only guess and I need my computer, I'll be more cautious.
ADDENDUM: "Earth, Air, Fire, Water." Today my RA is 75F, and my four-core maximum average for LinX fully-loaded temperatures is ~68C. So I made some slight improvements there, just for disabling PLL Overvoltage and reducing PLL voltage to 1.68V. Only time would tell if this slight improvement gets me lower load VCORE, but my last 15-pass sample tells me -- "Not much." Maybe next week I'll take another sample for giving it a closer look. Anyway -- all this occurs with a D14 cooler and a single, better fan.