2500k overclock VID Question

[zubbs1]

I've begun my manual overclock. Nothing big, just a 4.2 goal. I'm using:
Asus p8z68 pro gen 3 mobo
G.SKILL Sniper Low Voltage Series 240-Pin DDR3 SDRAM DDR3 1600 (1.25v)
Phantek PH-TC14PE 140mm Heatsink - model PH-TC14PE
Corsair TX850 V2 850W power supply

CPUZ
Prime95 v 27.7 - small fft
Real temp

CPU Current Capability - 140%
Phase and Duty Control - Extreme
EPU Power saving - Disabled
VRM Frequency - Manual - 350
LLC - Ultra (4/5)

I set the vcore to 1.25, run the prime 95 small fft test for several hours (while at work or asleep), then bump the vcore down.
I've now tested 1.25, 1.24, 1.23, and they are stable.

My question so far is about VID. In realtemp, regardless of what vcore I set, the VID shows 1.3811 under 100% load. The offset is obviously negative, but with it being that large, won't my system crash at idle, or whenever it ramps up to load? Also is VID suppose to scale with vcore maintaining the same delta?

 

[Jman13]

I ran at 1.23 for a good while at 4.2...no issues at all using offset. At 4.2 and 1.22 I had some minor issues.

 

[zubbs1]

I ran at 1.23 for a good while at 4.2...no issues at all using offset. At 4.2 and 1.22 I had some minor issues.

Do you remember what your offset was at either the 1.23 or 1.22 vcore?

 

[Kenmitch]

My question so far is about VID. In realtemp, regardless of what vcore I set, the VID shows 1.3811 under 100% load. The offset is obviously negative, but with it being that large, won't my system crash at idle, or whenever it ramps up to load? Also is VID suppose to scale with vcore maintaining the same delta?

There is no way to manipulate the VID voltage. It is coded into the chip and is based on multiplier being used, the load, and even the type of load and number of cores loaded. So it's basicaly the stock voltage for your chip at the ??x multiplier under your current load situation.

For clarification the VID is what intel told the chip it needs to be stable under the current speed and load. It's just a value and cannot be changed. It will always show in apps as what was hard coded into the chip.

 

[zubbs1]

There is no way to manipulate the VID voltage. It is coded into the chip and is based on multiplier being used, the load, and even the type of load and number of cores loaded. So it's basicaly the stock voltage for your chip at the ??x multiplier under your current load situation.

So would a large negative offset (around a -0.16ish or more) be an issue?

 

[Kenmitch]

So would a large negative offset (around a -0.16ish or more) be an issue?

If you're chip has no issue and is stable then not gonna be a problem at all.

You can do some quick testing with prime while trying to find vcore with the following settings. These are for 20-30 minute pre testing before longer prime runs. Not sure why it works but it does work and will bsod very quickly if vcore isn't close to what is gonna be required. You can also change the FFT values to 1344 as it hits just as hard.

Looking at your original post it looks like you looked at a guide for 4.5ghz or higher overclocking and dialed down the multiplier. Some of those settings shouldn't be required at such a low overclock.

CPU Current Capability - 140%
Phase and Duty Control - Extreme
EPU Power saving - Disabled
VRM Frequency - Manual - 350
LLC - Ultra (4/5)

The above should be good on the defaults of the motherboard. The LLC will fight you on the offset voltage. The higher the LLC the more - offset you'll need to use.

 

[zubbs1]

If you're chip has no issue and is stable then not gonna be a problem at all.

You can do some quick testing with prime while trying to find vcore with the following settings. These are for 20-30 minute pre testing before longer prime runs. Not sure why it works but it does work and will bsod very quickly if vcore isn't close to what is gonna be required. You can also change the FFT values to 1344 as it hits just as hard.

Looking at your original post it looks like you looked at a guide for 4.5ghz or higher overclocking and dialed down the multiplier. Some of those settings shouldn't be required at such a low overclock.

CPU Current Capability - 140%
Phase and Duty Control - Extreme
EPU Power saving - Disabled
VRM Frequency - Manual - 350
LLC - Ultra (4/5)

The above should be good on the defaults of the motherboard. The LLC will fight you on the offset voltage. The higher the LLC the more - offset you'll need to use.

Excellent, thanks for the advice!

 

[borisvodofsky]

Ok, I'm going to share with you My super secret bios settings. On account we have the exact same chip and board. :biggrin:

http://s1149.photobucket.com/albums/o583/borisvodofsky/

Minor changes I've made a month back, my (new settings)

CPU PLL: use 1.650v, more stable than 1.8125v

If you're going for 4.7ghz You can actually use Offset - 0.25 on LLC (HIGH)

That will solidly get you 4.7

Everything else is the same as in the picture. I also just got 2133 ram for kicks, but trust me, it does absolutely jack, don't bother.

You can change your Command Rate the 5th number in the dram timing to 1T if you are using 2 sticks. If you are using 4 sticks, change it to 2T


for 4.8+ this is where you may or may not need to go all the way to 1.4v

If your chip needs 1.4v, the voltage is fine, but I'd only recommend it if you have very good cooling, since it's summer.

During the winter, you can forget everything and use 1.4v without thinking about it.


As far as prime goes, you want to test 2 settings, 1792k, and 2688k, for 35 minutes each, and use 85% of your total memory.

so if you have 8gb total, you're gonna use ~6600


MAKE sure both those switches on the motherboard epu and tpu are turned OFF.

AND IF ANYONE tries to tell you to enable any of the C states, they're n00bs.

Turn off ALL C-states, they hamper SSD performance AND lowers stability.

 

[BonzaiDuck]

Ok, I'm going to share with you My super secret bios settings. On account we have the exact same chip and board. :biggrin:

. . . . . . . . . . .

Like I said . . . . I'm more cautious. I think I chose to stick at the second-lowest or 25% LLC setting. For this reason, I think my Offset voltage setting is slightly positive, while that of others here seems to be in the negative range. My load voltages flip between 1.31 and 1.33V. I thought I saw somewhere that you showed a 1.33V, but I was curious if that was the LOAD voltage.

Point being -- my idle voltage (said this many times before in other threads) briefly shows 1.36+V before dropping to EIST and 1.008V.

What you don't see with high LLC settings is an additional overshoot going from load to idle -- probably never even recorded in the monitoring software. But . . . that's a personal choice.

 

[MadScientist]

This maybe helpful. Idontcare's post in this thread: http://forums.anandtech.com/showthread.php?t=2194577&highlight=

As you rightly noted, offset is usually computed by taking the delta of the VID versus applied Vcc at full load.

But be aware that the way the offset method works is that the offset is applied regardless the VID. So the same offset is in effect when your VID changes to a lower value when the chip goes idle such that the formula "Vcc = VID + offset" is always in effect.

This means your idle volts are higher than they would otherwise be, maybe not a problem, but it can be a problem if your offset is actually negative.

My cpu was applying too much voltage at 4.5GHz, the VID was calling for more voltage than was necessary. So I found the minimum voltage for IBT stability, applied the negative offset and was happy as a lamb.

That is until I found out that the negative offset was actually pushing my idle volts so low that the rig was locking up when the chip idled.

So you really do need to ensure the offset you select is an offset that leaves your rig stable at both its loaded and unloaded extremes.

With a positive offset I don't see this ever being a problem, it only should be a problem if you are using negative offsets.

 

[zubbs1]

Thanks Boris for the help. I'll work on that in the coming days.

Another follow up question to anyone out there:

I set my vcore to 1.210. Under load CPUZ shows voltage from 1.208 to 1.216.

Can vdroop actually become anti droop? Why would it overshoot vcore with llc at 4/5?

 

[borisvodofsky]

Like I said . . . . I'm more cautious. I think I chose to stick at the second-lowest or 25% LLC setting. For this reason, I think my Offset voltage setting is slightly positive, while that of others here seems to be in the negative range. My load voltages flip between 1.31 and 1.33V. I thought I saw somewhere that you showed a 1.33V, but I was curious if that was the LOAD voltage.

Point being -- my idle voltage (said this many times before in other threads) briefly shows 1.36+V before dropping to EIST and 1.008V.

What you don't see with high LLC settings is an additional overshoot going from load to idle -- probably never even recorded in the monitoring software. But . . . that's a personal choice.

my 1.33v is during small ftt in PRIME95, If I use IBT, it will shoot to 1.35 sometimes even 1.36

This is because different loads at "the same multiplier, 48x" have different VID reported by the CPU, and the motherboard will responds to the VID.

I am using HIGH LLC, it's the "50%" setting w/ - 0.25

My processor requires a minimum of 1.33 to sustain 48x, if it ever dips below this value, it will become unstable.

So you need to tweak your voltage settings around this SPECIFIC LOAD, such that it will NEVER go below 1.33 even though you DON"T need 1.35 for IBT.

For example, I can tune it so that IBT passes at 1.33v maximum, and it'll be very cool while running IBT, however this setting will crash in Prime, because the lower vid reported during PRIME.

It pisses me off too.

 

[borisvodofsky]

Thanks Boris for the help. I'll work on that in the coming days.

Another follow up question to anyone out there:

I set my vcore to 1.210. Under load CPUZ shows voltage from 1.208 to 1.216.

Can vdroop actually become anti droop? Why would it overshoot vcore with llc at 4/5?

There are DIFFERENT vids reported by the CPU and the SAME multiplier at DIFFERENT cpu loads.

small ftt in prime reports a lower vid, than does intel burn test with avx. :biggrin:, i mean

 

[BonzaiDuck]

my 1.33v is during small ftt in PRIME95, If I use IBT, it will shoot to 1.35 sometimes even 1.36

This is because different loads at "the same multiplier, 48x" have different VID reported by the CPU, and the motherboard will responds to the VID.

I am using HIGH LLC, it's the "50%" setting w/ - 0.25

My processor requires a minimum of 1.33 to sustain 48x, if it ever dips below this value, it will become unstable.

So you need to tweak your voltage settings around this SPECIFIC LOAD, such that it will NEVER go below 1.33 even though you DON"T need 1.35 for IBT.

For example, I can tune it so that IBT passes at 1.33v maximum, and it'll be very cool while running IBT, however this setting will crash in Prime, because the lower vid reported during PRIME.

It pisses me off too.

If I recall, you were using same or similar ASUS mobo to mine. I had the first release of the P8Z68-V Pro board. So you are using a higher LLC setting than I. Also, your load voltages (for a 2500K) seem close if not identical to my load voltages for a 2600K. Perhaps there is some slight difference: I have HT ENABLED on my system. I might have disabled it after getting stability with HT, but didn't. Disabling it now affects my Media Center setup for my SiliconDust HDTV input. I think I'd have to call the cable provider again to reinitialize my cableCard.

But there is an instantaneous voltage overshoot between load and idle such that LLC will make the momentary spike more pronounced. This had been explained in an article here at Anandtech on OC'ing a QX processor at time of the Kentsfield or Yorkfield release (before Nehalem.) I don't think anything has changed with that. The authors of the article simply noted that you should "be aware of what is happening with LLC settings."

What you see as 1.35 to 1.36V or whatever you see while the processor is still at turbo speed but unloaded -- would be lower than the actual overshoot.

Anyway -- minor point, probably -- depending on the test [Prime95 FFT, LFT . . . IBT Max, High . . . etc.] I think my load voltages can drop as low as 1.31 to 1.32V with my 4.60+ speed options. In the end, it seems pretty consistent with yours except for your higher multi and my lower LLC.

Also, once I had stable settings at 4.65 and 4.7, I left the Offset and "Xtra Volts at Turbo" alone and dropped the speed to a flat 46 multi and default bCLK -- I'm actually slightly over-volted for this speed.

 

[borisvodofsky]

If I recall, you were using same or similar ASUS mobo to mine. I had the first release of the P8Z68-V Pro board. So you are using a higher LLC setting than I. Also, your load voltages (for a 2500K) seem close if not identical to my load voltages for a 2600K. Perhaps there is some slight difference: I have HT ENABLED on my system. I might have disabled it after getting stability with HT, but didn't. Disabling it now affects my Media Center setup for my SiliconDust HDTV input. I think I'd have to call the cable provider again to reinitialize my cableCard.

But there is an instantaneous voltage overshoot between load and idle such that LLC will make the momentary spike more pronounced. This had been explained in an article here at Anandtech on OC'ing a QX processor at time of the Kentsfield or Yorkfield release (before Nehalem.) I don't think anything has changed with that. The authors of the article simply noted that you should "be aware of what is happening with LLC settings."

What you see as 1.35 to 1.36V or whatever you see while the processor is still at turbo speed but unloaded -- would be lower than the actual overshoot.

Anyway -- minor point, probably -- depending on the test [Prime95 FFT, LFT . . . IBT Max, High . . . etc.] I think my load voltages can drop as low as 1.31 to 1.32V with my 4.60+ speed options. In the end, it seems pretty consistent with yours except for your higher multi and my lower LLC.

Also, once I had stable settings at 4.65 and 4.7, I left the Offset and "Xtra Volts at Turbo" alone and dropped the speed to a flat 46 multi and default bCLK -- I'm actually slightly over-volted for this speed.

Sry, Bonzai duck, I am having trouble following what you're trying to say?

But, if you're looking to get 48+ on a 2600k w/HT enabled you'll almost surely need at least 1.39v to 1.42v , MINIMUM.

so it must be at least 1.4 at low load @ 4.8ghz, like prime small ftt.

Under the same setting, it'll probably hit 1.45-1.48 during IBT, but that's not a problem since very few desktop apps actually loads up the cpu like IBT does.

the voltages I've reported are not voltage spikes, they're the long run numbers that is fed to the core.

 

[BonzaiDuck]

Sry, Bonzai duck, I am having trouble following what you're trying to say?

But, if you're looking to get 48+ on a 2600k w/HT enabled you'll almost surely need at least 1.39v to 1.42v , MINIMUM.

so it must be at least 1.4 at low load @ 4.8ghz, like prime small ftt.

Under the same setting, it'll probably hit 1.45-1.48 during IBT, but that's not a problem since very few desktop apps actually loads up the cpu like IBT does.

the voltages I've reported are not voltage spikes, they're the long run numbers that is fed to the core.

Most of the people I've crossed paths with here last year using similar mobos were either conservative as I am -- settling for around a 4.6 overclock, or they were pushing to the limit on water-cooling to 4.8 or 4.9. These were all 2600K users. There were maybe 10% of the 2600K retail-box CPUs that would get you to 4.9 or higher. The remainder were split, so you could push to 4.8 or so, or settle for 4.6 or 4.7.

So you're using a 2500K with HT Disabled? That must be very good. So I'll clarify again -- your LLC mitigates vDroop, which probably allows you to OC more. There is a split second when your processor unloads itself -- say, after a stress-test. This results in a voltage overshoot before reaching the "undrooped" level-- lower than the overshoot. But the LLC makes the overshoot more pronounced until settling to the "undrooped" level at that speed. The monitoring software will never show the spike -- you'll never see it.

You can visualize this as harmonic fluctuation that attenuates to an equilibrium level. Look around 2007 or 2008 at Anandtech for a paper here on a QX quad-core processor and over-clocking.

And like I said --they only noted that you have to be aware of what LLC does.

For me, under real stress as I said, my loaded voltage is somewhere between 1.31 to 1.33 depending on the test (Prime small-FFT versus IBT and the various levels within). When the test ends and the processor is unloaded at the top speed, it settles back to about 1.36+ before dropping to the EIST speed and voltage level. But with LLC enabled, the momentary spike (which I cannot see) may go as high as 1.4. But you don't know -- you don't know if it goes to 1.38 or 1.39 or even 1.41V. The monitoring software will never pick it up.

We're all gambling that (a) we have a decent idea of a "safe" range; (b) that the momentary fluctuations in voltage are of little consequence to the overall health of the chip. And (c) -- the processor spends most of its time at speed and voltage under the stock spec (3.4 for the 2600K), so it has less a chance to become degraded by electromigration over time.

 

[borisvodofsky]

Most of the people I've crossed paths with here last year using similar mobos were either conservative as I am -- settling for around a 4.6 overclock, or they were pushing to the limit on water-cooling to 4.8 or 4.9. These were all 2600K users. There were maybe 10% of the 2600K retail-box CPUs that would get you to 4.9 or higher. The remainder were split, so you could push to 4.8 or so, or settle for 4.6 or 4.7.

So you're using a 2500K with HT Disabled? That must be very good. So I'll clarify again -- your LLC mitigates vDroop, which probably allows you to OC more. There is a split second when your processor unloads itself -- say, after a stress-test. This results in a voltage overshoot before reaching the "undrooped" level-- lower than the overshoot. But the LLC makes the overshoot more pronounced until settling to the "undrooped" level at that speed. The monitoring software will never show the spike -- you'll never see it.

You can visualize this as harmonic fluctuation that attenuates to an equilibrium level. Look around 2007 or 2008 at Anandtech for a paper here on a QX quad-core processor and over-clocking.

And like I said --they only noted that you have to be aware of what LLC does.

For me, under real stress as I said, my loaded voltage is somewhere between 1.31 to 1.33 depending on the test (Prime small-FFT versus IBT and the various levels within). When the test ends and the processor is unloaded at the top speed, it settles back to about 1.36+ before dropping to the EIST speed and voltage level. But with LLC enabled, the momentary spike (which I cannot see) may go as high as 1.4. But you don't know -- you don't know if it goes to 1.38 or 1.39 or even 1.41V. The monitoring software will never pick it up.

We're all gambling that (a) we have a decent idea of a "safe" range; (b) that the momentary fluctuations in voltage are of little consequence to the overall health of the chip. And (c) -- the processor spends most of its time at speed and voltage under the stock spec (3.4 for the 2600K), so it has less a chance to become degraded by electromigration over time.

2500k doesn't have HT to begin with LOL.

I'm not pushing processors too hard these days. It spends most of it's time at 1.6ghz browsing the internet. ^_^

 

[BonzaiDuck]

2500k doesn't have HT to begin with LOL.

I'm not pushing processors too hard these days. It spends most of it's time at 1.6ghz browsing the internet. ^_^

Yeah -- I was aware of that -- altho it's easy to forget if you hadn't bought one and lived with a 2600K.

See, when push comes to shove, I only have couple games -- simulators, etc. Not Call of Duty or more recent and demanding entertainments. I do this on a lark, trying to keep abreast of a technology which now looks uncertain with "any mobile devices." On the other hand, I'm going to add more "duties" to the machine, so it not only gives me my home theater (which doesn't hardly put a dent in the CPU meter), but extends my "home security." I'll take my time with that. If I could get my computer to kill all the squirrels in my garden, I'd do it. But the more wonderful s*** you get your computer to do for you, the more careful one might be about choosing limits on the OC part.

Some of these folks would say "No! don't do it!" Well, 8 years, haven't lost anything.

 

[borisvodofsky]

Yeah -- I was aware of that -- altho it's easy to forget if you hadn't bought one and lived with a 2600K.

See, when push comes to shove, I only have couple games -- simulators, etc. Not Call of Duty or more recent and demanding entertainments. I do this on a lark, trying to keep abreast of a technology which now looks uncertain with "any mobile devices." On the other hand, I'm going to add more "duties" to the machine, so it not only gives me my home theater (which doesn't hardly put a dent in the CPU meter), but extends my "home security." I'll take my time with that. If I could get my computer to kill all the squirrels in my garden, I'd do it. But the more wonderful s*** you get your computer to do for you, the more careful one might be about choosing limits on the OC part.

Some of these folks would say "No! don't do it!" Well, 8 years, haven't lost anything.

Having read your posts, I've concluded you may in fact be more crazy than I am.

That, and I'm almost jealous.

 

[zubbs1]

Most of the people I've crossed paths with here last year using similar mobos were either conservative as I am -- settling for around a 4.6 overclock, or they were pushing to the limit on water-cooling to 4.8 or 4.9. These were all 2600K users. There were maybe 10% of the 2600K retail-box CPUs that would get you to 4.9 or higher. The remainder were split, so you could push to 4.8 or so, or settle for 4.6 or 4.7.

So you're using a 2500K with HT Disabled? That must be very good. So I'll clarify again -- your LLC mitigates vDroop, which probably allows you to OC more. There is a split second when your processor unloads itself -- say, after a stress-test. This results in a voltage overshoot before reaching the "undrooped" level-- lower than the overshoot. But the LLC makes the overshoot more pronounced until settling to the "undrooped" level at that speed. The monitoring software will never show the spike -- you'll never see it.

You can visualize this as harmonic fluctuation that attenuates to an equilibrium level. Look around 2007 or 2008 at Anandtech for a paper here on a QX quad-core processor and over-clocking.

And like I said --they only noted that you have to be aware of what LLC does.

For me, under real stress as I said, my loaded voltage is somewhere between 1.31 to 1.33 depending on the test (Prime small-FFT versus IBT and the various levels within). When the test ends and the processor is unloaded at the top speed, it settles back to about 1.36+ before dropping to the EIST speed and voltage level. But with LLC enabled, the momentary spike (which I cannot see) may go as high as 1.4. But you don't know -- you don't know if it goes to 1.38 or 1.39 or even 1.41V. The monitoring software will never pick it up.

We're all gambling that (a) we have a decent idea of a "safe" range; (b) that the momentary fluctuations in voltage are of little consequence to the overall health of the chip. And (c) -- the processor spends most of its time at speed and voltage under the stock spec (3.4 for the 2600K), so it has less a chance to become degraded by electromigration over time.

This made me curious:

Is there a software application that will run silently in the background, but track your cpu load over time, displaying a line graph to show when load occurs, and what programs were drawing that load? I'd be more agreeable to a bigger overclock if I had more knowledge of these parameters.

 

[Idontcare]

This maybe helpful. Idontcare's post in this thread: http://forums.anandtech.com/showthread.php?t=2194577&highlight=

Thanks for posting that :thumbsup: I couldn't remember where I posted it before but it was exactly what I knew the OP needed to know about.

 

[zubbs1]

Thanks for posting that :thumbsup: I couldn't remember where I posted it before but it was exactly what I knew the OP needed to know about.

So is it better to decrease the size of the negative offset (move it closer to zero) or decrease the level of llc (from 4.5 to 3 or 2/5) if you get the idle lock up or bsod issue?

 

[guskline]

I've begun my manual overclock. Nothing big, just a 4.2 goal. I'm using:
Asus p8z68 pro gen 3 mobo
G.SKILL Sniper Low Voltage Series 240-Pin DDR3 SDRAM DDR3 1600 (1.25v)
Phantek PH-TC14PE 140mm Heatsink - model PH-TC14PE
Corsair TX850 V2 850W power supply

CPUZ
Prime95 v 27.7 - small fft
Real temp

CPU Current Capability - 140%
Phase and Duty Control - Extreme
EPU Power saving - Disabled
VRM Frequency - Manual - 350
LLC - Ultra (4/5)

I set the vcore to 1.25, run the prime 95 small fft test for several hours (while at work or asleep), then bump the vcore down.
I've now tested 1.25, 1.24, 1.23, and they are stable.

My question so far is about VID. In realtemp, regardless of what vcore I set, the VID shows 1.3811 under 100% load. The offset is obviously negative, but with it being that large, won't my system crash at idle, or whenever it ramps up to load? Also is VID suppose to scale with vcore maintaining the same delta?

zubbs1: I have the same mb and just used the AIsuite OC software to get to a stable 4.5 Ghz. Much easier and runs well for months. Passes all the diagnostic tests (IntelBurn, Prime95, etc.).

 

[zubbs1]

zubbs1: I have the same mb and just used the AIsuite OC software to get to a stable 4.5 Ghz. Much easier and runs well for months. Passes all the diagnostic tests (IntelBurn, Prime95, etc.).

What does cpu-z show for your vcore under 100% load?

 

[BonzaiDuck]

This made me curious:

Is there a software application that will run silently in the background, but track your cpu load over time, displaying a line graph to show when load occurs, and what programs were drawing that load? I'd be more agreeable to a bigger overclock if I had more knowledge of these parameters.

I started using "Intel Turbo Boost Technology Monitor 2.0" -- which shows a small processor cap on your desktop and a bar graph of processor speed. I see it ramp up briefly for regular desktop apps, sometimes to levels lower than the 4.6+ turbo overclock I've set. Running my TV, web-browsing, reading PDF documents and so on -- the system pretty much remains at a 1.6+ Ghz EIST speed and voltage of 1.008V.

I personally don't think it's a good idea to run more than one monitoring program at a time. My ASUS AI Suite II monitor will show either the processor speed or the voltage, but not both at once. I haven't had any trouble using it to track voltages while the Intel Monitor is also running, however.

You can use several "disciplined" approaches. For instance, as you go through the over-clocking chores, you can stop after reaching a voltage equal to or only slightly greater than what stock "Turbo" (3.8Ghz for 2600K) will give you. At that point, with decent cooling, you might not even think about core temperatures. Or -- you can get opinions and form your own about what the upper limit of a "safe" range for the processor is or might be. That's basically what I did, and I limited use of LLC simply based on the Anandtech article I mentioned and the fact that "you don't know" empirical details about a voltage over-shoot. Or -- you can hang your a** over the edge a little bit and stop OC'ing when the load voltage itself is 5% above that limit. Or -- you can simply use the core temperatures as a limiting factor and "just try for a minimum voltage" with your result.

We had discussed this at length on threads last year. Based on the progression of processor "die-size" or the spec reported in nanometers, cores had gone from 45nm to 32nm and now to 22 nm. The Sandy Bridge is a 32nm core. They stopped including "safe range" in Intel specs after either Wolfdale/Yorkfield (45nm) or the first gen I7 Nehalems.

One forum member I remember noted that 1.30V was an upper limit on SB cores. Many more of us thought it was 1.35V. A veteran OC'er and WC-er acknowledged by many suggested that 5% overage was "nothing."

Another rule of thumb -- a 13% OC is harmless. IF you went from a Turbo of 3.8 to 4.3 Ghz, that's a 13% over-clock. I and most others have never reported or seen a processor have an untimely death with a 25% OC, and for the SB 2600K that would be in excess of 4.75Ghz.

I built a Q6600 system in midsummer 2007 @ 3.0Ghz (most would consider an easy or modest OC). That's a 25% over-clock. It's still running without fail or problem at my brother's house, and could probably last as long as the 10-year expected time for stock processors.