Safe maximum voltage? - Intel Vccin, Vss, Vcore

[Senzelian]

So it's 4 o' clock in the morning here in Germany and I'm wondering what the highest, but safe voltage on a Haswell-E CPU is.

We all know the answers: "Don't go above 1.4V, or the CPU dies". - "1.5V will burn your house down" - "1.6V will melt steel beams!" - right?

The thing is that I have never seen anyone doing an actual test about this - probably because no one wants to kill their CPU I guess.
So instead I thought I look a bit into Intel's "LGA2011-3 Socket Datasheet" in hope to get the final answer...

 

First you should know that this datasheed is probably the most confusing piece of text I've ever seen since the existance of the god damn bible, but that won't stop me from reading it.
Maybe at that point I should link to it: http://www.intel.com/content/www/us/en/processors/core/core-i7-lga2011-3-datasheet-vol-1.html

Anyway, I was looking around and found the "Processor Absolute Minimum and Maximum Ratings" on page 51.

Now there are different voltages listed, like Vccin, Vccd, Vccio_in and Vccpeci. And if you're confused at that point I can understand why.

So the first thing I did was to find out what all of these things actually mean. I found this very helpful site: 
http://www.hardwaresecrets.com/understanding-all-voltage-configurations-from-the-motherboard/4/

 

They wrote the following about Vcc:

Quote

VCC: This is the main CPU voltage, which can also be unofficially referred as Vcore. Usually, when we say “CPU voltage” we are talking about this voltage. The option that changes this voltage will show up on the motherboard setup as “CPU Voltage,” “CPU Core,” etc.

So if Vcc is the "main CPU voltage" or "Vcore" as we all know and love it, then it might get interesting now.

Intel claims that the absolute maximum Vccin (Vcc IN) voltage for 2011-3 CPUs is 1.98V.
Okay, done. Let's all overclock the hell out of our CPUs and put them at 1.98V!

Yeah, good luck with that. Intel also states, that these ratings are "outside functional operation condition limits, but within absolute maximum and minimum ratings"  and that "neither functionality nor long-term reliability can be expected". So that means, that even if you run your CPU at 1.98V it probably won't die, but you won't be able to get anything out of it other than a blackscreen.

But that leaves us with the question, where is the point that leaves us with a functional, stable but heavily overclocked CPU without risking the lifetime of it? 1.3V? 1.4V? 1.5V? 1.6V? 1.7V?
Well I don't know and that is where I have a ton of questions.

Who told you, that 1.4V is the maximum you should ever go for on a CPU?
By how much do we reduce the lifetime of our CPUs if we're going higher than that?
Is this even covered by warranty?
And the most important one: Am I even correct with all the assumtions I just made? (I probably did a ton of mistakes and at the end there is probably something wrong with it, but we'll see...)


 

Now it's 4:22am. Good night...
 

Edited March 31, 2016 by Senzelian

[runit3]

I killed my old 5960X with 1.78v Vcore under DICE (repeated runs, probably hit 1.85v at one point). VCCIN over 2.0v is not likely to gain you any more stability than leaving it at a respectable 1.85-1.9v (it's a moot point). If you really wanted to fry something pushing 1.5v+ through the IMC via VCCIO/SA is a nice way to kill the usability of a chip.

 

The new 5960X runs at 1.395v daily (under water) and I have had no degradation over the last 6 months. The only chip I've ever gotten to degrade was a 2550k at 1.525-1.5v for around a month of use (had to notch back the clock by 200MHz and eventually dropped it to 1.45v -no issues ATM reported from the person I gave the build to). This was also coupled with multiple sprints to 1.7v, so take that for what it's worth.

 

If you really want to get into the nitty-gritty of voltage limits no manufacturer flat out tells people where the maximums are. Mass trial and error reporting is really the only usable metrics we get, and outside of a few people on this forum who regularly push DICE/LN2, overclock.net is the place you will want to go for that type of information.

http://www.overclock.net/t/1510388/haswell-e-overclock-leaderboard-owners-club

 

For warranty info there's really no way to prove your CPU was running at a certain voltage, or for how long it was running that voltage. Many people regularly do warranty claims on chips they've killed with voltage, if there's no immediate physical signs of abuse (toasted pin platforms or deformed IHS's) there's not much they can do to refute the claim. This is obviously less than savory, but you can purchase Intel's PTPP for your chip ($25-35 depending on the chip) and blast it with voltage to your heart's content. This is what I did for my 5960X and told the warranty rep that I killed it under voltage. New chip arrived in less than 2 weeks (US shipping) -not sure if that plan is offered in Germany, or how long it would take to get a replacement.

 

 

[Prysin]

1.98v is probably the maximum limit of the conductive traces inside the CPU before the voltage is high enough for currents to jump over to other paths and cause a arc/short.

[Ansau]

- In Haswell, Haswell-E, Broadwell and Broadwell-E vccin is NOT vcore. Vccin is the voltage for the FIVR and vcore is the actual voltage going through the cores.

- There's no a true maximum safe voltage. Each cpu will degrade differently at a same voltage, even some will die with no apparent dangerous voltage. 

- The only way to prevent degradation is to have the components in a super-conductive state, which requires temperatures near 0 K. This is technically and practically impossible, so we can assume degradation is impossible to avoid. Even at idle at 0.1v the cpu is degrading (at a minuscule rate but it is still happening).

- Temperatures and, consequently, cooling solution are a key part to prevent degradation. 1.3v at 90ºC is more dangerous than 1.5v at 60ºC. Temperature affects directly and indirectly the electromigration. It is a divisor in the formula used to calculate the time to failure from electromigration and an exponent to calculate the conductivity of a material, which is used to know the current density which is also a divisor in that formula.

 

People are too scared about voltage and degradation. There's tons of people freaking out because you exceeded 1.35v, or people think degradation happens in a matter of months.

The true is that if proper cooled, latest cpus can handle high voltages without issues for extended periods of times, and the only true way to kill or severely degrade a cpu is to go insane with the voltage in air/water.

[tomaatvk]

I don't really think there exists such a specific voltage. No chip is perfect and any imperfection can affect the way it handles voltages, it may have a slightly higher/lower resistance at certain places, it may have transistors drain a little bit quicker or slower, it may have voltage jumps in the chip sooner or later. And not to mention, it may wear quicker or later. All depending on the specific piece of silicon you have.

For myself, I have a 3770k, it was used so I don't know how old it is, at what voltage it ran and how much it ran but I do know that I have been running it for around 8 months now and from the beginning I've ran it at a Vcore of 1.35v (according to my asrock z77 extreme 4 which is known to report too low voltages so it's probably more like 1.4v) and recently (the past 5 months or so) it has been running at roughly 1.4v according to my motherboard (wanted a higher overclock and 1.4v is probably more like 1.45v irl). I've also ran the thing at 1.5-1.6v quite a few times for several hours just to do some benchmarks at 5.1-5.3GHz (so that is probably 1.55-1.7v or so) and this was all on air (although the air was ~0°C) so some chips sure can take a beating. The CPU also didn't lose any stability, it still needs the same voltages as before to run at a specific speed.

On the other hand, I've heard of ivy bridge chips dying at around 1.55 or 1.6v and of some chips becoming less stable over time. An example of this I've heard is that a 3570k at first needed roughly 1.35v to run at something like 4.6GHz and after a year or so it needed 1.4+v for the same speed.

[Senzelian]

2 hours ago, Ansau said:

- In Haswell, Haswell-E, Broadwell and Broadwell-E vccin is NOT vcore. Vccin is the voltage for the FIVR and vcore is the actual voltage going through the cores.

This was very informative.
But what's FIVR? 

[runit3]

36 minutes ago, Senzelian said:

This was very informative.
But what's FIVR? 

Fully Integrated Voltage Regulator

http://hothardware.com/news/haswell-takes-a-major-step-forward-integrates-voltage-regulator

[Senzelian]

4 minutes ago, runit3 said:

Fully Integrated Voltage Regulator

http://hothardware.com/news/haswell-takes-a-major-step-forward-integrates-voltage-regulator

Ahhh...
Okay, well I know what that is. I just didn't know what the shortcut stands for.

[Masada02]

I'm just going to throw my personal experience in here for your reference.

 

I like to keep my 5960X under 1.3v for the daily OC, mainly because I like seeing a 1.2xx voltage value for a 4.6ghz OC. Depending on what OC your CPU can achieve, I'd go as far as 1.400v for daily use. When I am benching it will go as far as 1.580v depending on the bench I am running. I will not run 1.580v for a Cinebench run, but for many gaming benchmarks I will go that high. This is all on my EK Predator 360, not a custom loop or any really fancy cooling.