vdroop

[superstition]

If you've ever overclocked a system, chances are that at some point or another you've had opportunity to become upset with your Vdroop "problem." Some users, confused as to why their system refuses to exactly match actual processor supply voltage to the value specified in BIOS, are quick to blame the quality their motherboard; still others find fault with the difference noted between their board's idle and full-load processor supply voltages. Actually, load line droop (Vdroop) is an inherent part of any Intel power delivery design specification and serves an important role in maintaining system stability. In most cases, comments regarding unacceptable power delivery performance are completely unfounded. To make matters worse, unjustified negative consumer perception surrounding this often misunderstood design feature eventually forced a few motherboard manufacturers to respond to enthusiasts' demands for action by adding an option in their BIOS that effectively disables this important function.

Overclocking... The Rules Have Changed

How have the rules changed? vdroop certainly isn't a new thing for 45nm parts.

Doing this seems like a really bad idea to me. Vdroop is used to cut costs - basically, they can use less capacitance on the output of the voltage regulator and still supply "good enough" power to the CPU by simply overvolting it slightly when it's not drawing much current, and undervolting slightly when it's drawing a lot of current.

The naive reaction would be, "well, I'll disable this and always supply the right voltage". A user does that, and verifies with some voltage measuring tools (a multimeter, software, whatever) that the voltages look better. But the user missed what really changed. With active voltage positioning enabled, the transient spikes in the power supply are reduced but the steady-state voltages look worse. With it disabled, the steady-state voltages look better, but transient spikes will be worse (and AFAIK there is no way for software to detect the transients). If the CPU suddenly transitions from idle to load, there will be a brief moment when it is significantly undervolted (and, like any undervolted CPU, may crash / corrupt your data / whatever).

linear
thetechrepository

And yeah, don't mess with Vdroop. Doing the pencil mod or enabling Load Line Calibration and setting the voltage in BIOS lower, say from 1.412 to 1.345, isn't really going to help you much. Right when the load starts, the voltage will still dip to the same level as before, and then stabilize at just below the new now lower BIOS vcore of 1.345V, and when going from load to idle, it will shoot back up to your old BIOS vcore of 1.412V, possibly higher, and then stabilize at 1.345V again. This takes only milliseconds, not enough for any software to measure, but in those milliseconds your cpu can crash. So don't get rid of Vdroop. If you need higher load voltage, just increase vcore in BIOS. The vcore set in BIOS is just the maximum specified voltage, not the actually vcore you run anyways.

"Vdroop" exists because Intel decided that it's needed. There was at least this article here at anandtech that explained voltage dropping under load:
Everything I say in following assumes that the charts in this article are correct, and show exactly the idea of what happens.

So "Vdroop" is a GOOD thing, because it reduces the variance in processor vcore. It is also a bad thing for those that care more about their average vcore than their absolute maximum vcore value at some short moment. Personally I'd rather have less vcore variance and a bit higher average vcore than more variance and lower average.

I think it goes like this:
Too little voltage - and you experience errors in CPU work. That means instability. But I doubt it damages the hardware. (just software, in a way)
Too much voltage - and you shorter the lifespan of your CPU. I do not know the details. This can damage hardware. (who needs software after that?)

The optimal amount of vdroop is a mystery though. Too much vdroop and the variance is again greater than it would be without vdroop. Perhaps the ideal value depends on motherboard and processor used? Never trust your motherboard manufacturer though, and it wouldn't surprise me a bit that your mobo manufacturer (or maybe their BIOS team) just puts some value there, which ends up being not-so-ideal. Hopefully some other poster knows more abut this.

Forum people that say "motherboard A is better than B, because A has less vdroop" annoy me a bit... at least after reading this anandtech article. For overclockers who aren't ready to break their hardware, the following question is relevant: "Will running my CPU with a higher average vcore cause me more trouble (that can be a shorter CPU lifetime or lower achievable OC or marginally higher power usage) than running my CPU with a lower average vcore, but higher occasional vcore?" (that occurs when CPU load drops). Of course assuming that the Vdroop solution in question does what it is meant to do....

EDIT: Original Poster... do you have C1E and/or EIST enabled in BIOS?? Vdroop is about the voltage drop that occurs when your CPU is *under load*. The power saving mechanisms will probably both lower your CPU multiplier to 6 and also lower your vcore if your processor is at *idle state*!

Disable the power saving options to see if your vcore is the same than what you set it to be. After that run some CPU stressing application and during the run, open CPU-Z and check your vcore to see the "Vdroop".

So, are people going to continue doing pencil mods?

 

[JAG87]

Thanks for quoting me

Vdroop is a good thing, people dont seem to understand that. Intel specifies vdroop to be +/- 0.1v, yet when people see 0.1v of droop they freak out. As a matter of fact, anything less than 0.1v of droop is at fault. Ok, not entirely... read on

The longer version of that: some manufacturers use far better mosfets and power regulation on their board (asus, gigabyte, DFI), and can afford to make the droop smaller because their voltage regulation is far more accurate and less spiky. Intel's spec is made with regular 4 or 6 phase power regulation systems in mind.

So people, if your motherboard droops 0.1v under load, its not shit. It's just made with cheaper power delivery cuircuit design and cheaper components. Next time buy an 8 phase 400 dollar Asus motherboard, and you wont have as much vdroop. The safe way, not the buy-a-cheap-motherboard and pencil-mod way.

 

[bryanW1995]

I am once again very upset that I didn't get the DQ6...

 

[nevbie]

Thanks for quoting me

err.. i mean..

It is said that motherboard sensors and stuff aren't to be trusted. For example processor temp and various voltages. But does a typical motherboard have the correct scale in their vcore measurement, or is that often flawed too?

If the scale is OK but the values are wrong, then it would still be enough to see if vdroop is close to that 0,1v or whatever is the ideal value. It might be better to undervolt and turn vdroop off if the drop is too large. I guess the answer is no though, and motherboard measurements are not to be trusted. =P

If the voltage regulation itself can be too "spikey", then there isn't much to say about this subject though. At least if that cannot be seen from vcore readings.

I suppose there are always some risks in overclocking and that's all.

 

[superstition]

So, who still feels pencil mods (or similar - like that Asus bios setting) are a good idea?

 

[Idontcare]

Originally posted by: superstition
So, who still feels pencil mods (or similar - like that Asus bios setting) are a good idea?

What's lacking in this debate is the viewpoint of the other side of the argument.

I.e. who has done a pencil mod and realized a benefit to it?

If there are no realized benefits, say improved overclocking or reduced temps at load or something...then its a no-brainer that its at best a pointless idea and at worst causes the harm that Anandtech and others have well documented.

I have never bothered to dip into the pencil-mod threads so I have no idea what the allure is...but if we don't hear from anyone on the matter then you can't really have a debate or a conclusion.

On the topic of loadline calibration I can add a data point. With my QX6700 under phase change cooling on an Asus P5E WS Pro I had to sock 1.67V to the chip (1.675V BIOS, 1.65V at idle) just so it would survive the Vdroop to 1.5V under load at 4GHz for small FFT stability.

1.67V did not make me comfortable as an idle voltage. With loadline calibration I set the Vcore in BIOS to 1.53125V, it runs 1.528V at idle in XP (I like 1.53V vs 1.67V for idle hours) and when fully loaded with small FFT the Vcore moves up slightly to 1.560V thanks to loadline calibration.

So obviously the question is "how big is the unload transient"? Is it bigger than 0.11V? (1.67-1.56 = 0.11V) If it isn't bigger than that then clearly loadline calibration is a good thing for my case.

But even if it is bigger than 0.11V the question really ought to be framed "which does the least amount of aggregate damage over time - a chip experiencing idle voltages of 1.67V for hours at a time every day when my computer is not loaded, or a chip experiencing 1.67V (or 1.7V or 1.75V...) only for 10ms or so during the unload transients which occur infrequently?"

For the duration of the time I had my chip operating in that fashion I felt the loadline calibration was a net plus for my already tortured chip.

 

[rge]

On my Gigabyte P35 DQ6 board, I have seen no benefit to loadline with a dual core. But on that board, vdrop and vdroop are small anyways. With my E8400, with loadline DISABLED, Vdrop(.03) + vdroop(.01) are ~.04V. To get 4.1 stable, I need 1.375V in bios, 1.344 CPUZ idle, and drops~.01v on load. With loadline ENABLED, to get 4.1 stable, I need 1.34 bios, 1.34cpuz idle, and stays 1.34 load. One notch below in vcore in bios of either of the above two settings, and I fail prime 95, at both above I am stable 12+ hours. And if loadline increases the overshoot spikes, all I am getting in return is at most .01v less at idle (mathematically most possible based on sensor only reads changes in steps of .016V) and load is either near same or slightly more.

I think in my setup, its a no-brainer to leave loadline disabled. Also that article referenced above is simply going to keep me buying quality motherboards that have low vdrop/vdroop and still in specs. But in Idontcare's example above, I would have to pull out the multimeter, and measure the overspikes versus idle versus load, and make a determination based on that.

 

[PCTC2]

Yeah. I don't believe in pencil-mods. I have a 680i with 6-phase power. I used to be one of the ignorant of vDroop. but with the amounts of voltage i used to push through my chips, i got 0.1v vDroop. but definitely no on the pencil mods. The PWM circuitry cannot instantaneously control voltage. Switching from load to idle, the total wattage stays the same for that split moment that it takes for the PWM to react, and due to laws of physics and electrics, when the amperage drops at a constant wattage, the voltage peaks, and that can cause damage, unneeded wear and tear, or instability in your system.

 

[superstition]

Evidently, this isn't the final word because I've read two topics in Motherboards within the last 24 hours where people were talking about load line calibration and pencil mods.

 

[aigomorla]

Originally posted by: superstition


So, are people going to continue doing pencil mods?

Sigh... Jag can i just came out of a debate on this. This is so YMMV, you need to look at what your doing and why your doing it.

Also to answer your question:

Yes... and i will be one of them. If you understand what your doing then moding the vdroop is safer then not doing it.

Example. you have a chip that loads @ 1.45Vcore. You know this because you tested it. Your board has a nasty vdroop of .1 or greater. IE IP35-E is one board, Some of the ASUS would be another 680i/780i

I have a thing where i dont like any voltge over 1.5. So my only fix, is trying to lower that idle voltage.

According to anand's article, the spikes and browns are bad. However the spikes and browns wont put that voltage anywhere near 1.6. If i didnt pencil mod i would hit 1.6. The spikes will actually be somewhere close to 1.5 but not exceed it. Problem now is brown, however if your chip can handle that brown off load thats fine as well.

Pencil mod good or bad?


Each case is a YMMV tag. You cant just dismiss a trick without knowing why it was really implimented. And how to pull it off safely.

You wouldnt point a loaded gun at someone without knowing which direction it fired. Same concept.

 

[Markfw]

I like my DQ6, and it works better than any of my other motherboards. AND IT HAS NO VDROOP.

I agree to disagree.

 

[core2slow]

I'm still very noobish in trying to understand about the whole over/undervoltage, vdroop, pencil mod and their relationships to CPU.

So lets say i set my BIOS at 1.45v
at idle im 1.38v (per cpuz)
at load im 1.32v (per cpuz)

So therefore, my vdroop is 0.13v (1.45v-1.32v)? Since it's greater than 0.1v, i have a crappy mobo, correct?

 

[lopri]

I also wanted a definite answer but so far it's still murky at best. I do understand that vDroop is a result of inevitable law of physics, however:

1. 1) vDroop specified by Intel does NOT assume overclocking.
   
   2) My understanding is that Intel's guideline is about 'allowable limits' rather than a 'recommendation' or 'mandatory'.
   
   3) Accepting vDroop as a preferable occurence fail to explain the varying degree of vDroops observed on different build quality of motherboards. (Note the build quality doesn't always translate to less/more vDroop, but there are limits of each component)
   
   4) Revisiting #1, say, let's take an example of an overclocked CPU which requires 1.4V to maintain its stability. Which would be a better situation, having to give 1.5V idle 90% of the time in order to secure 1.4V load 10% of the time or having a solid 1.4V which might overshoot to 1.45V under load?

I am not outright against more knowledgeable folks advice, but there are things that I haven't been answered satisfactorily.

 

[superstition]

The higher the voltage, the greater the vdroop, as far as I've read.

 

[lopri]

Originally posted by: superstition
The higher the voltage, the greater the vdroop, as far as I've read.

That is the law of physics part.

 

[Idontcare]

Originally posted by: lopri
I also wanted a definite answer but so far it's still murky at best. I do understand that vDroop is a result of inevitable law of physics, however:

4) Revisiting #1, say, let's take an example of an overclocked CPU which requires 1.4V to maintain its stability. Which would be a better situation, having to give 1.5V idle 90% of the time in order to secure 1.4V load 10% of the time or having a solid 1.4V which might overshoot to 1.45V under load?[/list]

To be sure if the transient voltage spike is lower than the steady-state voltage at idle then you have made a net gain for your CPU.

The real quandry comes when you have a spike (say 1.55V) that is higher than the steady-state (1.5V) as then the question of frequency versus duration comes into play.

For a given length of time you might run the chip exposed to 1.5V there will always be an equivalent amount of damage you would have done if the idle voltage was <1.5V but the transients are >1.5V. The higher the transients then obviously the lower the frequency of transient events you can subject your chip to and have the equivalent damage.

But you really need to know the trade-offs on chip lifetime that each option causes before you can be justified in assuming one or the other is superior for a given computer usage pattern.

 

[Markfw]

What I know, is that the better motherboards overclock more, are more stable and have less vdroop.

So in the overclockers world, its bad IMO....I have the experience and the boards to prove that.

 

[Tweakin]

Originally posted by: JAG87

So people, if your motherboard droops 0.1v under load, its not shit. It's just made with cheaper power delivery cuircuit design and cheaper components. Next time buy an 8 phase 400 dollar Asus motherboard, and you wont have as much vdroop. The safe way, not the buy-a-cheap-motherboard and pencil-mod way.

So it's better to buy a $400 dollar board to take your chip to 4GHz then it is to by a inexpensive board at $75 that only takes you to 3.8? I always thought overclocking was getting the most for the least...unless you are trying to set records.

 

[JAG87]

Originally posted by: Tweakin

Originally posted by: JAG87

So people, if your motherboard droops 0.1v under load, its not shit. It's just made with cheaper power delivery cuircuit design and cheaper components. Next time buy an 8 phase 400 dollar Asus motherboard, and you wont have as much vdroop. The safe way, not the buy-a-cheap-motherboard and pencil-mod way.

So it's better to buy a $400 dollar board to take your chip to 4GHz then it is to by a inexpensive board at $75 that only takes you to 3.8? I always thought overclocking was getting the most for the least...unless you are trying to set records.

All im saying is don't whine about it.

If your motherboard has vdroop, its there for a reason. It's calibrated according to how good or poor the VRM is on the board. Example: Mark has zero vdroop on a gigabyte DQ6 board with 12 phase VRM, as opposed to core2slow who has 0.13v vdroop on a DS3L with 4 phase VRM.

Great we discovered hot water :roll:

 

[Idontcare]

Originally posted by: Tweakin
So it's better to buy a $400 dollar board to take your chip to 4GHz then it is to by a inexpensive board at $75 that only takes you to 3.8? I always thought overclocking was getting the most for the least...unless you are trying to set records.

It really depends on your specific overclocking or underclocking objectives, naturally.

If you wander over the XS forums you'll see that they recognize this and they break-out all the varying hardware aspects to overclocking feats.

It very much becomes a "one size fits one" situation. Getting a $400 mobo to reach 4GHz clocks is no "better" than getting a $75 mobo to reach 3.8GHz clocks...unless your specific overclocking desire needs a 4GHz clock to be satisfied.

Back when quads first came out, Nov 2006, I was intent on hitting 4GHz with a B3 Kenstfield so I bought the $1500 QX6700, the $500 mobo, and the $1000 vaporphase LS needed to hits those clocks for 24/7 stable operation back then.

It certainly was not the "best" way to overclock, it all came at a price premium, but for my purposes if I wasn't going all the way at any price then it was pointless to spend anything less if I got less of an overclock.

Nowadays I love my DS3L's that get me comfy but slower overclocks in the 3.3GHz range. Those systems serve a different purpose where price/performance was critical.

 

[Tweakin]

Originally posted by: JAG87

Originally posted by: Tweakin

Originally posted by: JAG87

So people, if your motherboard droops 0.1v under load, its not shit. It's just made with cheaper power delivery cuircuit design and cheaper components. Next time buy an 8 phase 400 dollar Asus motherboard, and you wont have as much vdroop. The safe way, not the buy-a-cheap-motherboard and pencil-mod way.

So it's better to buy a $400 dollar board to take your chip to 4GHz then it is to by a inexpensive board at $75 that only takes you to 3.8? I always thought overclocking was getting the most for the least...unless you are trying to set records.

All im saying is don't whine about it.

If your motherboard has vdroop, its there for a reason. It's calibrated according to how good or poor the VRM is on the board. Example: Mark has zero vdroop on a gigabyte DQ6 board with 12 phase VRM, as opposed to core2slow who has 0.13v vdroop on a DS3L with 4 phase VRM.

Great we discovered hot water :roll:

Then I misunderstood you...I agree. Most don't need to worry about the vdroop on their boards.

 

[Tweakin]

Originally posted by: Idontcare

Originally posted by: Tweakin
So it's better to buy a $400 dollar board to take your chip to 4GHz then it is to by a inexpensive board at $75 that only takes you to 3.8? I always thought overclocking was getting the most for the least...unless you are trying to set records.

It really depends on your specific overclocking or underclocking objectives, naturally.

If you wander over the XS forums you'll see that they recognize this and they break-out all the varying hardware aspects to overclocking feats.

It very much becomes a "one size fits one" situation. Getting a $400 mobo to reach 4GHz clocks is no "better" than getting a $75 mobo to reach 3.8GHz clocks...unless your specific overclocking desire needs a 4GHz clock to be satisfied.

Back when quads first came out, Nov 2006, I was intent on hitting 4GHz with a B3 Kenstfield so I bought the $1500 QX6700, the $500 mobo, and the $1000 vaporphase LS needed to hits those clocks for 24/7 stable operation back then.

It certainly was not the "best" way to overclock, it all came at a price premium, but for my purposes if I wasn't going all the way at any price then it was pointless to spend anything less if I got less of an overclock.

Nowadays I love my DS3L's that get me comfy but slower overclocks in the 3.3GHz range. Those systems serve a different purpose where price/performance was critical.

That was the point I was trying to make. Unless you need to reach for the gold, for whatever reason, a few precious MHz will not make the difference in the outcome. If it's for the e-penis award of the year, break out the bankroll as you are going to need it. If you want a nice, snappy machine and are trying to save a few hundred bucks by not getting the latest-greatest highest cost processor, then go with a nice board and chip and save your money for a vacation to Disney...or a vaporchill unit

 

[superstition]

tcsenter

Since vDroop is programmable (to a degree), it really depends on implementation and tuning more than anything. Solid caps make it worse, all other things equal, because the entire purpose of vDroop is to get away with using less bulk capacitance. Though 'solid' caps offer lower ESR over a wider temperature range, they also have less capacitance than good electrolytic caps of equal dimensions, and cost nearly twice as much to boot.

As a general proposition, it should be true that more output phases results in lower vDroop, it really doesn't work out that way in practice. Its not hard to find four phase designs with lower vDroop than motherboards with more costly six and eight phase designs. ASUS and Gigabyte have produced boards with really classy power supply designs that should be keeping vDroop to a best case minimum, but are no better or even worse than some 'cheaper' boards.

kjboughton

Vroop has nothing to do with the number of power delivery phases.

Intel power delivery design specifications include a requirement to maintain "load line resistance" within a small window. Some motherboard manufactures may choose to ignore these requirements to some degree or may lack the ability to control overall build tolerances to the extent necessary to limit vdroop to allowable levels. As such some boards, even runs withing a particular product line, may have more or less vdroop than others.

This has nothing to do with the quality and/or capabilities of a circuit. If the system is unable to provide enough power one of two things will happen:

1) one of more power delivery components will most likely overheat and fail, or
2) higher than expected current levels will drag down the controller and the system will shutdown when the programmable VID leaves the allowable regulation range (spontaneous reset).

The bottom line? - The level of vdroop is not a proper method for comparing two power delivery systems.

 

[Markfw]

and a DQ6 is not $400, like I paid $225 for 2 of mine and $186 for the 3rd. And any decent motherboard is ~$100. I paid $60 for my IP35-E,and it doesn't do near as well as my $130 DS3R.

 

[LOUISSSSS]

Originally posted by: Markfw900
and a DQ6 is not $400, like I paid $225 for 2 of mine and $186 for the 3rd. And any decent motherboard is ~$100. I paid $60 for my IP35-E,and it doesn't do near as well as my $130 DS3R.

yea i've been hearing that the ip-35 doesn't get as good of an overclock as the ds3 series...
at least its cheap. i wouldn't mind getting one for $60