Only my opinion but I think ~1.4V max if you want no damage to your CPU and ~1.52V max if you have adequate cooling and don't mind the CPU loosing a couple hundred Mhz off it's max overclock over time.
I folded my 2600K for 1.5 months last year at 4.8Ghz at 1.47V, under water, temps were something like 65C starting out. It did lose stability at the end, but it's hard to tell what other factors were in play. The radiator had gotten full of dust and the ambient temp had started rising as summer rolled in and I didn't pay close attention to core temp.
I think electromigration does start occuring past ~1.4V, from my experience and reading tons of forum posts of Sandy Bridge. The rate it happens appears dependent on temperature. I personally think under 1.52V at a reasonable distance from Tcase max for 24/7 will not outright "kill it", although I would say your max overclock probably drops off 100Mhz-200Mhz after a few month or so. More importantly if you return it to stock speeds after 1.52V 24/7 it will probably last as long as you want to keep the processor.
However, my personal stance for a 24/7 machine... I have done competitions and pushed voltage and temps envelopes for a point here or a point there, but 100% 24/7 is an entirely different animal. Most importantly is recognizing diminishing returns. To make up some numbers: is 200Mhz worth 1.46V over 1.38V and 10C of temperature difference, how much degradation are you submitting the CPU to for 200Mhz? In folding that nets you maybe 1000 more PPD over the guy that used the Intel stock cooler. What if you spend $200 for watercooling to gain a couple hundred Mhz, your half way to a low end pieces parts i5 system that will net you 10,000 or more PPD. That said I think for a 24/7 machine that is going to be at 100% CPU it's best to just stick with 1.4V max. Give your overclock leeway and not riding the limit of stability on a 24/7 rig.
I haven't played with it in so long but try this experiment. Each CPU has it's own VID's set at sort and test, but in the BIOS/EFI try bypassing overcurrent protections but leave voltage and turbo voltage offsets at 0, dont touch loadline calibration, and set a 45 or 46 multiplier, then run a non AVX app at 100% and look at requested VID, then run a AVX app(LinX or something) and look at requested VID. I may be completely wrong it was over a year ago but for some reason I thought I saw 1.38V and 1.42V(AVX), which would be how intel designed it. The only thing you are variable you are intervening in is telling the CPU you have the heat removal capability to increase the stock limits for current.
I folded my 2600K for 1.5 months last year at 4.8Ghz at 1.47V, under water, temps were something like 65C starting out. It did lose stability at the end, but it's hard to tell what other factors were in play. The radiator had gotten full of dust and the ambient temp had started rising as summer rolled in and I didn't pay close attention to core temp.
I think electromigration does start occuring past ~1.4V, from my experience and reading tons of forum posts of Sandy Bridge. The rate it happens appears dependent on temperature. I personally think under 1.52V at a reasonable distance from Tcase max for 24/7 will not outright "kill it", although I would say your max overclock probably drops off 100Mhz-200Mhz after a few month or so. More importantly if you return it to stock speeds after 1.52V 24/7 it will probably last as long as you want to keep the processor.
However, my personal stance for a 24/7 machine... I have done competitions and pushed voltage and temps envelopes for a point here or a point there, but 100% 24/7 is an entirely different animal. Most importantly is recognizing diminishing returns. To make up some numbers: is 200Mhz worth 1.46V over 1.38V and 10C of temperature difference, how much degradation are you submitting the CPU to for 200Mhz? In folding that nets you maybe 1000 more PPD over the guy that used the Intel stock cooler. What if you spend $200 for watercooling to gain a couple hundred Mhz, your half way to a low end pieces parts i5 system that will net you 10,000 or more PPD. That said I think for a 24/7 machine that is going to be at 100% CPU it's best to just stick with 1.4V max. Give your overclock leeway and not riding the limit of stability on a 24/7 rig.
I haven't played with it in so long but try this experiment. Each CPU has it's own VID's set at sort and test, but in the BIOS/EFI try bypassing overcurrent protections but leave voltage and turbo voltage offsets at 0, dont touch loadline calibration, and set a 45 or 46 multiplier, then run a non AVX app at 100% and look at requested VID, then run a AVX app(LinX or something) and look at requested VID. I may be completely wrong it was over a year ago but for some reason I thought I saw 1.38V and 1.42V(AVX), which would be how intel designed it. The only thing you are variable you are intervening in is telling the CPU you have the heat removal capability to increase the stock limits for current.
Last edited:
