Disabling C1E, C3 & C6 causes voltage increase in CPU-Z

[Danny55]

Hi Guys,

I've got a 2500k and a ASRock P67 Extreme4 ()

When I disable C1E, C3 and C6 to constantly run @ 3300MHz I get a voltage increase in CPU-Z, 1.232V instead of 1.200V when everything is enabled.

Although my UEFI reports 1.200V CPU-Z reports 1.232V, this also happens when I'm overclocking, for example 4.5GHz @ 1.304V in UEFI is shown as 1.363V in CPU-Z.

So what does this mean? Am I actually running @ 1.232V or just 1.200V?

Which one should I trust? And why does CPU-Z report this? :'(

 

[veri745]

It probably has to do with either the VRM (load line) or CPU-z inaccuracy. I wouldn't worry about it too much, are you having any other symptoms?

 

[Danny55]

Well actually I never kept it on this setting cuz I was scared that I would overvoltage my CPU.

Do you think it's safe to run a stresstest on these settings?
And do you think that I can assume that I'm running the voltage which the UEFI is telling me?

My Load Line is currently on Auto (Level 5) if you woud like to know.

Edit: And btw when the C settings are enabled the voltage in CPU-Z is the same as in the UEFI.

 

[Idontcare]

Which one should I trust? And why does CPU-Z report this? :'(

You can't trust either. If you really want to know the voltage on the CPU you have to physically measure it with a voltmeter.

Some mobos like the Asus extreme IV Z68 have the voltage points made easy to access for measuring, others are much more difficult to access.

The bottom line is that in all likelihood both the UEFI and CPUz are misreporting the cpu voltage. They are close, like you car's gas gauge, but you can't expect them to be anything but ballpark correct.

 

[Danny55]

Ok but why does my UEFI and CPU-Z report the same and good voltage when all C states are enabled?

Cuz this only happens when I turn off the C states.

But lets say if my system is running 4.5GHz stable when I set the VCore to 1.302V, will it still run stable after I turned off the C states and CPU-Z is reporting a higher voltage?

 

[Danny55]

Ok now this is awkward, I just started Prime95 and my voltage went from 1.232 down to 1.184V according to CPU-Z.

 

[mrSHEiK124]

Ok now this is awkward, I just started Prime95 and my voltage went from 1.232 down to 1.184V according to CPU-Z.

This is called VDroop - you need to do some reading :\

 

[Rvenger]

Totally normal voltage fluctuations. Mine does the same thing. If I run prime 95 my voltages will stick at 1.27v. When I run a simple WinRAR benchmark test, it will spike up to 1.32v. The cpu pretty much uses what it needs but it will wall out at Intel spec.]

Basically if your voltage doesn't spike beyond 1.35v, I wouldn't worry. Anything below that is fine, especially if you are not overclocking.

 

[BonzaiDuck]

Totally normal voltage fluctuations. Mine does the same thing. If I run prime 95 my voltages will stick at 1.27v. When I run a simple WinRAR benchmark test, it will spike up to 1.32v. The cpu pretty much uses what it needs but it will wall out at Intel spec.]

Basically if your voltage doesn't spike beyond 1.35v, I wouldn't worry. Anything below that is fine, especially if you are not overclocking.

Yes, it seems that 1.35V is as much recommended than my own initial choice to keep 32nm Sandy Bridge silicon within 32nm Nehalem specs.

And by this we mean an idle voltage of 1.35V [to 1.37V ?] This "idle" voltage may show up with some monitoring software that records Min and Max values, while the "drooped" "load" value will seem to prevail throughout stress-tests. Before the Sandy Bridge returns from its OC'd Turbo level to either the stock speed or the EIST speed, the processor may be "unloaded" for a short time before dropping speed and voltage.

These idle spikes are probably not too significant, and any harmonic peaks above what is recorded may not matter much as well. But it would be a mistake to volt the processor so that the drooped (load) voltage records at 1.35, 1.37 or higher.

Also, the fact remains that under ordinary usage, the voltage may not droop very much, so the higher voltage may prevail.

 

[Danny55]

Ok thx,

I won't worry about it anymore then

I am overclocking btw so do you recommand Fixed or Offset voltage for static speeds?

 

[BonzaiDuck]

Ok thx,

I won't worry about it anymore then

I am overclocking btw so do you recommand Fixed or Offset voltage for static speeds?

My understanding is that the P67 chipset -- to which Z68 is a hybrid successor -- has the same features for "Turbo" overclocking. On the ASUS boards, I have it from ASUS tech-support that they won't implement both base-clock over-clocking and Turbo over-clocking. I wouldn't be familiar with your options for the ASRock P67 board, but colleagues here have shown me screenies of the ASRock Z68 BIOS.

Some of us feel like Mark Twain must have felt about the passing of the Mississippi steamboat era: We may think that the over-clocking scenarios with Sandy Bridge have "taken the fun out" of over-clocking. While I experienced the same sentiment, I've come to a different view.

It is now possible and quite easy to perform the over-clocking with EIST, C1E, etc. enabled, keeping the base-clock speed (i.e., 3.4 Ghz for the i7-2600K), and overclocking the Turbo feature. In this scenario, VCORE and clock speed rise to the demand of processor load. This means that under ordinary operating conditions with low-load software, the average voltage is closer to the EIST voltage. This saves power, and should save the processor.

I'm guessing that the same features were implemented for P67 as for Z68: You can "fix" the VCORE, or leave it set to "Auto." There is an "Offset" voltage which can be adjusted, and finally, there is a feature which I call "VXtra" -- "extra voltage applied to VCORE in Turbo Mode." On the Z68 motherboard BIOS, these latter two can be adjusted while leaving the VCORE set to "Auto."

Add to that the LLC feature, which likely has four or five different settings or levels on either the ASUS or the ASRock motherboards. This then gives multiple options for attaining voltage and system stability on these boards with Sandy Bridge.

If Offset remains positive and the low-load Turbo voltage is at or just a little above the threshold we discussed, the harmonic spike following a load condition may be slightly more pronounced and exceed the target-voltage [which is the low-load Turbo voltage when speed is still highest.] If higher levels of LLC are applied, this spike may be even higher.

This was the point emphasized in an Anandtech article published here in December 2007, pertaining to the over-clocking of the QX9650 processor (LGA_775). All those parameters and considerations apply to Nehalem or Sandy Bridge, insofar as they address VCORE and its components (vOffset, vDroop, etc.)

With that in mind, you may want to try and keep low-load "Turbo-speed" voltage within -- say -- 1.37V. But consider that the processor is not always running at top speed or the highest voltage. When it is running at those levels, the load may be relatively light compared to a stress-test situation (when vDroop will be more pronounced unless it is countered by a higher LLC setting).

After experimenting with my Z68 build, I have to recommend over-clocking the "Turbo" Mode. This means that neither voltage nor speed are static. It also means less energy consumption. And -- with these chipsets and motherboards -- it seems to be quite stable.

So you would leave VCORE set to "Auto." "Offset" simply moves the voltage range up or down in increments of millivolts for all speeds -- from the base-clock to the top-end Turbo clock. You're recommended to minimize the Offset setting (which can be negative or positive). You can fiddle with LLC settings, but you should be aware of how higher LLC or higher Offset will generate a momentary spike in Load-to-Idle transition which cannot be measured. Any minor risks this might present are reduced if you can keep the low-load "idle" voltage at the Turbo speed below the "safe" limit for the processor.

This "safe" limit has not been published as a spec for Sandy Bridge yet by Intel -- for whatever reason they choose not to. But the SB processor is 32nm silicon as are later incarnations of Nehalem -- which has a "safe-range" limit of about 1.37V.

Many people here will probably tell you that following "rules-of-thumb" within these kinds of guidelines will certainly preserve the CPU's longevity; others will say the approach is "too sissy" and that you can push VCORE (either variable "auto" or "fixed") to higher levels.

You can choose your own "rules-of-thumb" or follow those recommended by any number of sources. I and some others here are just one such source.

 

[james1701]

You can't trust either. If you really want to know the voltage on the CPU you have to physically measure it with a voltmeter.

Some mobos like the Asus extreme IV Z68 have the voltage points made easy to access for measuring, others are much more difficult to access.

The bottom line is that in all likelihood both the UEFI and CPUz are misreporting the cpu voltage. They are close, like you car's gas gauge, but you can't expect them to be anything but ballpark correct.

Come to think of it, is there any way we could get someone to make a post with pictures measure voltages, and showing the check points, and how to set the meter to do it. That would make a good sticky topic. I don't think I have seen a full article on how to do that. It would be good for novices.

 

[BonzaiDuck]

Come to think of it, is there any way we could get someone to make a post with pictures measure voltages, and showing the check points, and how to set the meter to do it. That would make a good sticky topic. I don't think I have seen a full article on how to do that. It would be good for novices.

Unflattering -- for me to think of myself as a novice, but this would be a good idea. I've always been reticent about putting a multi-meter to my motherboard. Just read a post at Overclockers.com by "BobNova" offering a less-ominous view of LLC and unmeasureable, momentary load-to-idle-transition voltage spikes:

http://www.overclockers.com/load-line-calibration/

If anything, exploring a variety of views on this topic is still of value because it provides a sense of uncertainty/risk for what we "don't know" or can't measure easily.

 

[Bill Brasky]

More reading:

http://www.anandtech.com/show/2404/5

 

[BonzaiDuck]

More reading:

http://www.anandtech.com/show/2404/5

Thanks for posting that -- I refer to it often and keep a copy in PDF form to consult about these things.

 

[Idontcare]

Come to think of it, is there any way we could get someone to make a post with pictures measure voltages, and showing the check points, and how to set the meter to do it. That would make a good sticky topic. I don't think I have seen a full article on how to do that. It would be good for novices.

I don't mind doing this, I just bought an ASUS maximus iv extreme-z and this is one of the features touted for it.

Let me see what I can work up.

Here's a pic of the probe points:

 

[james1701]

I don't mind doing this, I just bought an ASUS maximus iv extreme-z and this is one of the features touted for it.

Let me see what I can work up.

Here's a pic of the probe points:

That would be great. I have the same belt on my Asus RE 3. For some reason, if you have ever taken a multi meter that was not set correctly and stuck it in an electrical outlet, and watched the sparks fly, there is no way someone would ever get one around their mobo. Not saying I have done that kind of thing , but its just a write up I have never seen any other site do. The correct way to test it, and what the entire belt on these new boards are used for, and how to do it.

 

[BonzaiDuck]

NOW! I can SEE what I might pay for in a $360 motherboard . . . . What a revelation . . .

On the upside to favor standing pat at this point, the UEFI-BIOS on my P8Z68-V-Pro gives me a sense of confidence in whatever monitoring potential the board exhibits. I can't make up my mind as to whether I should've waited --to look more carefully at a developing market.

By the way . . . . If Idontcare is going to work something up, some ASUS Maximus Manual excerpts would also be interesting . . .

I don't mind doing this, I just bought an ASUS maximus iv extreme-z and this is one of the features touted for it.

Let me see what I can work up.

Here's a pic of the probe points:

 

[Idontcare]

FWIW here's some data from notes.

Initially I was using an older BIOS (20X where X was either a 1, 2, or 3, I failed to document this) and the voltage differential between the BIOS and the multimeter was around 0.03V (sizable). A BIOS reading of 1.248V would actually be 1.218 on the multi-meter.

Then I updated the BIOS to ver 403, the latest one, and the BIOS voltages (and CPUZ) match the multi-meter readings out to the third decimal place (+/- 0.001V).

What this tells me is that there is no rule of thumb to be applied to the accuracy or inaccuracy of the BIOS/CPUZ voltage readings and the actual applied voltage because it can vary on the same mobo all depending on the BIOS settings created by the manufacturer.

In one case I had wildly inaccurate reported Vcc's from the BIOS and CPUZ, then a BIOS update later and the reported values are spot-on.

The board also consistently over-volts the dram by 0.021V. The BIOS reports a 1.501V Vdimm but the multimeter measures a consistent 1.522V Vdimm.

Not enough to be concerned about at these voltages, but my dimms are 1.5V rated (not 1.65V) and the mobo was automatically setting the Vdimm to 1.65V (Bios) which is not good for the CPU and not good for the dram either, but it wasn't 1.65V as it was actually further overvolting the dram to 1.671V per the multi-meter.

For a $360 mobo, cost more than my friggen CPU , I expected a lot more "intelligent" default settings from this thing. It does deliver in terms of options for configuring and optimizing your rig, but the defaults just seem to suck.

 

[sm625]

My company makes a 16 channel analog input PCI board that I could connect all 8 of those points to and monitor all of those voltages in software. It is a 16 bit ADC and factory calibrated to be accurate within 300 uV. Can take simultaneous samples at 100khz. It would be interesting to have that data. Would be great for power supply analysis.

Anyway, I've been unhappy with the performance of CPU-Z. It seems to stop working properly after a while. Like after minimizing more than a couple times. I have to restart it to get the right readings. I am looking for a better program, if there is one.

 

[GammaLaser]

The board also consistently over-volts the dram by 0.021V. The BIOS reports a 1.501V Vdimm but the multimeter measures a consistent 1.522V Vdimm.

I wonder if this is the board trying to account for a possible voltage drop going into the DRAMs if the DRAM power consumption is high, and then the BIOS is estimating the voltage at the DIMMs for this case