4.7 ghz vs 4.8 ghz

[MrDudeMan]

I am kind of hoping there is something useful and simple you can tell us. Ever since Intel stopped giving us safeish voltages we have been totally in the dark and unable to really say anything about safe voltages. Obviously its complicated and confidential but its not like we need the process testing details or the design schematics. We just want to set basic limits like 1.3V so that its unlikely that there will be a problem.

That is how we frame it, how do you as an expert in the process and these concerns distill down to the parameters that we have? We are looking for advice not the word of Intel. We don't need official, we just want some guidance.

I definitely understand your point and frustration. As another Intel engineer on these boards told me, citing numbers can be dangerous. The reason it's dangerous is because it gets propagated around as "some Intel guy said X volts is okay" and it becomes the golden standard. That can be bad for the career of the person who released a number (people have been fired over such things) and it's bad for the person who fried their CPU because they got a part that had less tolerance to overvolting. Yes, they could have fried it anyway, but they may have been less likely to try to squeeze the last 25mV if someone on the internet hadn't told them it was safe. See this link for an example. "...degradation starts..." is exactly the problem. I assure you it's happening way before 1.3V because it's already happening at the nominal voltage. I'm not trying to weasel out of providing an answer by saying it's complex because it truly is.

There was actually an internal article about what not to post on public forums this week. It's ironic, actually, because I don't usually post things like this at all.

If your goal is to squeeze every ounce of performance out of your CPU with power and reliability be damned, then it probably shouldn't bother you to go well above the nominal voltage.

If you want a modest overclock with some priority placed on power usage, voltage is important. The amount of energy required for every charge cycle is 1/2 * C * V^2 (C = capacitance, V = voltage). You are going to burn a lot more power for every incremental voltage increase due to how many transitions are happening across the entire die even with power gating. Energy turns into heat because of non-zero resistance and some, though very small, amount of complex impedance.

The best guidance I can give you is to figure out how much overclock you really want. I would then iteratively reduce the voltage as low as possible until instability is introduced. Then increase it until the instability goes away. I know, that's obvious, but it's an algorithm for people to follow who may be new to this.

For temperature, I would do my best to stay well under Tjmax in all operating conditions. Remember the ambient air temperature could fluctuate a lot and cooling solutions are only as good as the ambient temperature will allow. It's also worth noting that the thermal diodes don't always directly measure the die temperature. Some (many) transistors experience higher temperatures than the number that's reported to you.

For voltage, two things are important. If you exceed the dielectric strength of the gate oxide, it's game over probably immediately. If you get close to, but don't exceed, the breakdown voltage, it's probably still game over in a short period of time unfortunately. I actually don't know what the breakdown voltage is on the IVB or HSW processes, so I couldn't tell it to you anyway. Assuming the voltage is less than the breakdown voltage, then it becomes a complex relationship between increased heat and increased kinetic energy in the electrons flowing through the transistors, wires, and vias.

The final thing I'll say is with respect to the OP:

4.8 with 1.389 vcore and gaming temps of 66c.

I'm not giving any kind of official opinion by saying this. Personally, 1.389V seems incredibly aggressive and I would never do that to my own CPU.

 

[pandemonium]

A solidly, tactful response. Your job is safe!

 

[BrightCandle]

One of things we often see is that the next step of an over clock requires an ever increasing step in voltage. There certainly is a point (like in the ops case) where substantial voltage was needed to get it stable at the next step.

Is it fair to say that this is likely indicative of the fact that its pushed to far? Once we get away from the near linear relationship we are likely in very unsafe territory?

I personally don't like to push into or through these barriers, its not that I have broken a CPU that way but its hard to achieve 100% reliability when the clock speed was difficult to achieve to begin with. In the ops case I would argue this was hard fought stability and hence its not worth the effort, because its likely still not stable, just nearly stable and quite high voltage.

The other aspect of all of this is I would very much like to find a better stability test, I do wonder if anyone *cough* has some ideas of programs that validate a wider variety of instructions.

 

[2is]

One of things we often see is that the next step of an over clock requires an ever increasing step in voltage. There certainly is a point (like in the ops case) where substantial voltage was needed to get it stable at the next step.

Is it fair to say that this is likely indicative of the fact that its pushed to far? Once we get away from the near linear relationship we are likely in very unsafe territory?

I personally don't like to push into or through these barriers, its not that I have broken a CPU that way but its hard to achieve 100% reliability when the clock speed was difficult to achieve to begin with. In the ops case I would argue this was hard fought stability and hence its not worth the effort, because its likely still not stable, just nearly stable and quite high voltage.

The other aspect of all of this is I would very much like to find a better stability test, I do wonder if anyone *cough* has some ideas of programs that validate a wider variety of instructions.

It's quite easy to tell when you're forcing a CPU to run at a speed it clearly isn't comfortable running, and anytime I've done that, I always needed to turn the OC down later on in its life to maintain stability.

 

[MrDudeMan]

One of things we often see is that the next step of an over clock requires an ever increasing step in voltage. There certainly is a point (like in the ops case) where substantial voltage was needed to get it stable at the next step.

Is it fair to say that this is likely indicative of the fact that its pushed to far? Once we get away from the near linear relationship we are likely in very unsafe territory?

I think that's a fair assumption. When the gains come at a much higher price, it's because there are a lot of factors at play that have started encroaching on each other.

I personally don't like to push into or through these barriers, its not that I have broken a CPU that way but its hard to achieve 100% reliability when the clock speed was difficult to achieve to begin with. In the ops case I would argue this was hard fought stability and hence its not worth the effort, because its likely still not stable, just nearly stable and quite high voltage.

That's usually how I do it as well. I overclock as high as the platform can handle simply to see what the ceiling is, stable or not. Then I back it down until the performance gain is steep compared to the voltage increase.

I created the following chart to illustrate the point. The data points and shape are fictitious, so don't draw any conclusions from it in terms of real numbers. The point is to graphically explain diminishing returns for people who aren't familiar with the concept. The way to read the graph is to use the Y axis as frequency and the X axis as voltage. Let Y = 0 mean stock frequency and Y = 1 mean maximum overclock. The X axis numbers are equally spaced steps in VID (I didn't assign real values to the X axis on purpose).

At the beginning of the curve, slight increases in voltage yield substantial frequency increases. As you climb higher on the graph, more voltage is required to continue gaining frequency. Eventually the curve starts to lay flat, which means huge increases in voltage are required for minor increases in frequency. I'm more comfortable at or below the "knee" of the curve when I overclock. On this chart, the knee is approximately X = 6.

The other aspect of all of this is I would very much like to find a better stability test, I do wonder if anyone *cough* has some ideas of programs that validate a wider variety of instructions.

The programs everyone is already using are the only ones I know of in the public domain. I think they do a pretty good job based on personal overclocking experience. IDC's guide is probably as good as it gets without a lot more effort.

 

[WaTaGuMp]

Out of boredom, I did epeen flex to 5.0. Only ran Prime for 5 minutes though. One core hit 89c, too high for my taste. 1.45v bios, CPUZ had it at 1.42v under load with vdroop.

 

[Face2Face]

Out of boredom, I did epeen flex to 5.0. Only ran Prime for 5 minutes though. One core hit 89c, too high for my taste. 1.45v bios, CPUZ had it at 1.42v under load with vdroop.

Wow! That's quite the voltage jump to hit 5.0Ghz - Looks like 4.8Ghz on your CPU is the sweet spot. Same here pretty much. I can hit 5.0Ghz @ 1.356 - Not sure how stable it really is (Ran an excel bench with it). But to go from 4.8-5.0 is not worth the risk with the increased voltage.

 

[jacktesterson]

100 MHz

Try to see a difference.

 

[WaTaGuMp]

Wow! That's quite the voltage jump to hit 5.0Ghz - Looks like 4.8Ghz on your CPU is the sweet spot. Same here pretty much. I can hit 5.0Ghz @ 1.356 - Not sure how stable it really is (Ran an excel bench with it). But to go from 4.8-5.0 is not worth the risk with the increased voltage.

I just played around with it a bit trying lower voltage. Didn't have it stable, tried 1.40 and 1.42 with vdroop both were in the 1.37 and 1.38 range. I could most likely get it stable, but I would also not feel good unless I have a custom loop for cooling, or at the least a nicer kit. I have a good chip with great speed already, the rest is just bragging rights, heh. Someone on another forum with better cooling says he gets 70-75c with 1.47v. Are the better water cooling loops really that much better for temps, me at 89c and him maxed at 74c. My cooling is no slouch, but that's a pretty big difference in temps.

 

[guskline]

MrDudeMan is trying to tell the OCs (including me) something with his posts without getting in trouble with his employer. Read his posts carefully. Read IdontCare's posts carefully. What I come away with is keep the OC @4.3-4.5 MAX and watch that vcore voltage and temps. Intel has it's own extreme tuning software that is downloadable. I used it and found that though my 4.5 OC was stable the temps were a bit high for me. Backed it to 4.4Ghz 24/7 with Auto vcore. I tweeked the BIOS settings to allow more cpu power (130%?) and adjusted the LLC. Moreover, I ran OCCT for 1 hr, Intel stability test in their own software and carefully monitored temps.

I found 4.4Ghz was as high as I would go. If you want 4.7 to 4.8 on the Ivy Bridge you can probably get there but at what cost? High heat and with increased voltage likely damage to the cpu. Why for bragging points?

My 4.4Ghz OC is a rocket ship that is stable, cool and powerful.

 

[Face2Face]

MrDudeMan is trying to tell the OCs (including me) something with his posts without getting in trouble with his employer. Read his posts carefully. Read IdontCare's posts carefully. What I come away with is keep the OC @4.3-4.5 MAX and watch that vcore voltage and temps. Intel has it's own extreme tuning software that is downloadable. I used it and found that though my 4.5 OC was stable the temps were a bit high for me. Backed it to 4.4Ghz 24/7 with Auto vcore. I tweeked the BIOS settings to allow more cpu power (130%?) and adjusted the LLC. Moreover, I ran OCCT for 1 hr, Intel stability test in their own software and carefully monitored temps.

I found 4.4Ghz was as high as I would go. If you want 4.7 to 4.8 on the Ivy Bridge you can probably get there but at what cost? High heat and with increased voltage likely damage to the cpu. Why for bragging points?

My 4.4Ghz OC is a rocket ship that is stable, cool and powerful.

Come on buddy, live a little Everyone's chip is a different. If you could do 4.8Ghz @ 1.3v would you do it? Also how long do you plan on owning your CPU? With the rate I go through CPU's I couldn't imagine having this longer than 2 years.

 

[guskline]

It's not a matter of running the cpu at 4.8Ghz. It's whether or not it is stable at that speed and voltage. What tests have you run on it?

 

[MrDudeMan]

It's not a matter of running the cpu at 4.8Ghz. It's whether or not it is stable at that speed and voltage. What tests have you run on it?

Frequency is important. CMOS Dynamic power: P = C*V^2*f. It's linear with respect to frequency, so that's helpful in this context, but it's still in the equation. The CPU will consume ~6% more power running at 4.8GHz compared to 4.5GHz . That's a gross oversimplification, but it's probably close as a first order approximation.

 

[WaTaGuMp]

MrDudeMan is trying to tell the OCs (including me) something with his posts without getting in trouble with his employer. Read his posts carefully. Read IdontCare's posts carefully. What I come away with is keep the OC @4.3-4.5 MAX and watch that vcore voltage and temps. Intel has it's own extreme tuning software that is downloadable. I used it and found that though my 4.5 OC was stable the temps were a bit high for me. Backed it to 4.4Ghz 24/7 with Auto vcore. I tweeked the BIOS settings to allow more cpu power (130%?) and adjusted the LLC. Moreover, I ran OCCT for 1 hr, Intel stability test in their own software and carefully monitored temps.

I found 4.4Ghz was as high as I would go. If you want 4.7 to 4.8 on the Ivy Bridge you can probably get there but at what cost? High heat and with increased voltage likely damage to the cpu. Why for bragging points?

My 4.4Ghz OC is a rocket ship that is stable, cool and powerful.

I have never had a CPU die due to heat or voltage ever. These IB chips can run in the 70-80's all day long without issue. TJ max being at 105, with proper cooling there is virtually no risk for anyone who knows what they are doing.

 

[MrDudeMan]

I have never had a CPU die due to heat or voltage ever. These IB chips can run in the 70-80's all day long without issue. TJ max being at 105, with proper cooling there is virtually no risk for anyone who knows what they are doing.

That just means you aren't trying hard enough.

 

[Face2Face]

It's not a matter of running the cpu at 4.8Ghz. It's whether or not it is stable at that speed and voltage. What tests have you run on it?

It is for me and my uses. I have used Prime95 (6 hours) LinX, IBT and a couple others. I am using voltage offset along with Turbo. So every power saving feature is still enabled in the bios and when I am at idle the volts dip and so does the clock speed. I started out using this PC for gaming only, but now I am doing some Photo and Video editing. So far so good

That just means you aren't trying hard enough.

 

[WaTaGuMp]

That just means you aren't trying hard enough.

This is one instance where I don't think I will try hard enough. :biggrin:

 

[Tsaar]

Some anecdotal data:

I got a terrible 2600K last year. It takes me 1.4V with a really high LLC setting to get 4.5. My temps are OK with the Noctua (60ish peak). I did start around 1.385V last year and had to bump up to 1.4V once it started to get warm again this year.

I run 24/7 and it has been over a year now. I bought the Intel Tuning Plan just to be safe, but hoping I never have to use it.