Load-Line Calibration: why overclockers should care

[Godlygamer23]

-snip-

Higher phase count doesn't allow a similar overclock at a lower voltage. What a higher phase count does is allow for a more stable current to be fed into the CPU, GPU, or memory.

[MRAScott]

Higher phase count doesn't allow a similar overclock at a lower voltage. What a higher phase count does is allow for a more stable current to be fed into the CPU, GPU, or memory.

 

Perhaps. But the implication would be that the added cost of the VRM's (and the HS's used on them, sometimes even heat-pipe) on high-end MB's serve absolutely no purpose at all, except to sell higher-priced 'premium' product.

But anyway, isn't a more stable voltage/current capability precisely what stable overclocking (or underclocking/undervolting, for that matter) requires?

Whatever, coincidentally or not, the no-load/load-to-idle freezing/BSOD issue with negative offset vcore overclocks (1155 socket) was something that seemed to affect particular MB's, and IIRC they were 'budget' or mid-range editions,

[Godlygamer23]

Perhaps. But the implication would be that the added cost of the VRM's (and the HS's used on them, sometimes even heat-pipe) on high-end MB's serve absolutely no purpose at all, except to sell higher-priced 'premium' product.

But anyway, isn't a more stable voltage/current capability precisely what stable overclocking (or underclocking/undervolting, for that matter) requires?

Whatever, coincidentally or not, the no-load/load-to-idle freezing/BSOD issue with negative offset vcore overclocks (1155 socket) was something that seemed to affect particular MB's, and IIRC they were 'budget' or mid-range editions,

Having a more stable current doesn't mean you're going to have lower voltage. Power phases are called that because they work one at a time. That is to say, they go through phases. Normally you would have two or three transistors per choke. One choke may be rated at 25A. That means they all must be because only one set of transistors and one choke are working at any given time. The reason why we add more power phases is to give more stable current to hardware. Your claim was that a higher phase count gives a lower voltage for a particular overclock, which isn't(or at least shouldn't be) true. It'll provide a more stable overclock at least.

[MRAScott]

Having a more stable current doesn't mean you're going to have lower voltage. Power phases are called that because they work one at a time. That is to say, they go through phases. Normally you would have two or three transistors per choke. One choke may be rated at 25A. That means they all must be because only one set of transistors and one choke are working at any given time. The reason why we add more power phases is to give more stable current to hardware. Your claim was that a higher phase count gives a lower voltage for a particular overclock, which isn't(or at least shouldn't be) true. It'll provide a more stable overclock at least.

You may well be absolutely correct, I'm simply speculating about this (not actually making claims) on the basis of a couple of oddities I and others have experienced (with 67/68/77 intel chipsets and sandy/ivy bridge CPU's). BTW, I was spurred to search around a bit more, and some apparently solved the issue (with neg offset vcore and LLC) by changing various 'C state' settings, which I gather are concerned with power saving.

[Queek]

Hi - informative stuff, thank you.

*snip*

 

One caveat with using 'high' or above LLC settings, though;

one can end up with well under 1.0V vcore at idle, which can cause lock-ups or BSOD's when the machine is, er, idling. I've experienced this myself, despite the machine being overnight stable with Prime95 (or any other stress test) at 4.4Ghz with vcore at load (again, actual) of 1.34V.

I currently use medium LLC and a +ve CPU voltage offset of 0.020V, previously I was using a negative offset of 0.010V and high LCC ('high' appears to add another 0.04V at load compared to the medium setting), but idle voltage would sometimes drop below 0.9v and I had a couple of BSOD's when Windows (7 Pro 64) was loading, or when the PC was just clocking down from a stress test.

 

cheers,

Mark.

Thanks, I'm glad people are still reading this!

With adaptive overclocks used by more recent boards, one can have voltage drop below that required for stable operation at lower multiplier levels.  In my understanding, this does not have much to do with LLC, which appears to govern the stability of the voltage applied (what you set vs what you get).  The adaptive voltage curves are set by the motherboard manufacturer to represent the majority of cpus (as I understand it, citation needed), so if someone runs into instability at lower clocks, I usually recommend increasing the offset, as you said.  Doing this solved the problem on my Z97 overclock, and has solved the problem for several other people that I've recommend it to.  I don't think that this is a vdroop effect (LLC), and is instead a consequence of the general voltage curve determined by the manufacturer.  

[sgloux3470]

One might argue that higher levels of LLC make sense for people who want to run their CPU at lower voltages in idle and then ramp up under load.

[Legitsu]

idle is not the cause of instability 

its the fact that it doesn't add enough voltage fast enough when stepping up to a higher pstate/load from idle ... 

if your chip is decent it should't be having issues idling at 0.800 or less

example if the cpu is running at 800Mhz @ 0.750v @ 10% load if I open notepad there will be a moment where it will be 800Mhz @ >80% load before it steps up to the next frequency/voltage table

this is where your instablity will showup remember voltage required scales with load not Strictly Clock Speed

the solution is to increase the base voltage offset whilst using it in conjunction with the turbo voltage offset 

LCC can be used to help compensate for this but its not a real fix 

the real solution is for Microsoft to improve the windows cpu governor 

[MRAScott]

[Queek said]

 

"Thanks, I'm glad people are still reading this!

With adaptive overclocks used by more recent boards, one can have voltage drop below that required for stable operation at lower multiplier levels.  In my understanding, this does not have much to do with LLC, which appears to govern the stability of the voltage applied (what you set vs what you get).  The adaptive voltage curves are set by the motherboard manufacturer to represent the majority of cpus (as I understand it, citation needed), so if someone runs into instability at lower clocks, I usually recommend increasing the offset, as you said.  Doing this solved the problem on my Z97 overclock, and has solved the problem for several other people that I've recommend it to.  I don't think that this is a vdroop effect (LLC), and is instead a consequence of the general voltage curve determined by the manufacturer. "

 

 

.........

 

Absolutely - it does seem to be negative offsets that cause these 'wobbles' at idle with otherwise rock-solid machines, using higher than default LLC to achieve stability at high O/C's, LLC simply adhering to it's algorithm.

 

I noticed that if I open CPU / MB monitoring software as soon as I can into Windows boot (usually CPU-Z), the CPU is running at its (overclocked) multiplier for at least a minute, presumably while all the services and B/G processes intitialise, but Vcore is at barely over a volt, which might explain the early BSOD's with a  -ve offset driving it even lower (<0.9V).

 

IOW, perhaps asking a 3.3GHz CPU to run at 4.4GHz or more with such minute voltages, at whatever load, might be asking for a few errors?

 

Many of the reports of random 'idle' BSOD's were when people fire up something like a web-browser - again, despite having got their PC's completely stable in all stress tests (and demanding apps and games) -  at high O/C's. Momentary lag in Vcore when the multiplier shoots up for second or two?

 

(ETA >> did something when I was typing that screwed up the 'quote' formatting, hence the edit.

[Jumper118]

i always use extreme

[Queek]

MRAScott Many of the reports of random 'idle' BSOD's were when people fire up something like a web-browser - again, despite having got their PC's completely stable in all stress tests (and demanding apps and games) -  at high O/C's. Momentary lag in Vcore when the multiplier shoots up for second or two?

 

I can't figure out how to get that to quote properly.  

The crashing at idle may have something to do with the pre-programmed voltage/multiplier curve, which can be fixed with a slight offset (I wrote a thread about it earlier today here).  If it crashes going from idle to load, then it's probably an LLC problem; if it crashes randomly while idling, it's probably an offset problem.  

[MRAScott]

a few numbers to share;

currently my machine, an Asus P8Z77-V / i5 2500K  is @ 4.5 Ghz, Vcore offset +0.050V, LLC medium ("25%").

Constant full-load Vcore is 1.360 - 1.368V (both CPU-Z and Asus monitor).

Starting apps from idle, CPU-Z reports brief Vcore of 1.392V.

I don't ever want to see peaks going over that voltage with my i5 2500K, which is specified to run @ c.1.20V at it's standard clock of 3.3 GHz.

Remember;  these high-voltage "events" are whenever ANY application is started, which for me (and most people) will be TENS of 1000's of times a year.

Volts are like water-pressure - enough pressure, enough times, and a pipe will burst.
 

[Queek]

a few numbers to share;

currently my machine, an Asus P8Z77-V / i5 2500K  is @ 4.5 Ghz, Vcore offset +0.050V, LLC medium ("25%").

Constant full-load Vcore is 1.360 - 1.368V (both CPU-Z and Asus monitor).

Starting apps from idle, CPU-Z reports brief Vcore of 1.392V.

I don't ever want to see peaks going over that voltage with my i5 2500K, which is specified to run @ c.1.20V at it's standard clock of 3.3 GHz.

Remember;  these high-voltage "events" are whenever ANY application is started, which for me (and most people) will be TENS of 1000's of times a year.

Volts are like water-pressure - enough pressure, enough times, and a pipe will burst.

 

Software is notoriously bad at monitoring voltage, particularly at times of change.  Many programs are reasonably good at monitoring load voltages, but in my experience at idle or moderate usage, voltages are way off what my digital multimeter says.  That may be true voltage spikes, but I have a suspicious that it's just an artifact of the software.  

That is a reasonably high offset as well, which may be contributing to your spikes (from what I've heard, offset overclocking is not recommended).  A medium LLC on ASUS boards tends to be pretty conservative on the few boards that I've tested, usually resulting in a vdroop, not a vboost.  Is it possible for you to check your voltages with a DMM?

[MRAScott]

Software is notoriously bad at monitoring voltage, particularly at times of change.  Many programs are reasonably good at monitoring load voltages, but in my experience at idle or moderate usage, voltages are way off what my digital multimeter says.  That may be true voltage spikes, but I have a suspicious that it's just an artifact of the software.  

That is a reasonably high offset as well, which may be contributing to your spikes (from what I've heard, offset overclocking is not recommended).  A medium LLC on ASUS boards tends to be pretty conservative on the few boards that I've tested, usually resulting in a vdroop, not a vboost.  Is it possible for you to check your voltages with a DMM?

Funnily enough, I've had my DMM at the ready, set to 2000mV scale, but for the life of me I can't find where to measure Vcore on the P8Z77-V. Some Asus MB's have easily accessable, clearly marked measurement contacts, but not mine

 

it might well be that I'd have to go to the back of the board, which would neccessitate taking it out of the case and 'breadboarding' it.

[MRAScott]

BTW - prior to using isopropyl, use lighter-petrol.

[Queek]

Funnily enough, I've had my DMM at the ready, set to 2000mV scale, but for the life of me I can't find where to measure Vcore on the P8Z77-V. Some Asus MB's have easily accessable, clearly marked measurement contacts, but not mine

 

it might well be that I'd have to go to the back of the board, which would neccessitate taking it out of the case and 'breadboarding' it.

Probably not worth the effort then.  

If you're really worried that it's LLC causing your problem, try lowering the LLC and see if it lowers that voltage spike.  Lowering LLC may destabilize your overclock, though.  

 

 

BTW - prior to using isopropyl, use lighter-petrol.

What? No! Why would you say that? There's a specific entry in my thread saying why that's a bad idea and how it's unnecessary.  

[MRAScott]

Lighter petrol for cleaning heat-sink sufaces, not as a general cleaning agent for electrical contacts, I should clarify.

[sgloux3470]

Lighter petrol for cleaning heat-sink sufaces, not as a general cleaning agent for electrical contacts, I should clarify.

 

There's no benefit to this at all.  isopropyl alcohol is a much more effective cleaning solvent, as butane will evaporate very quickly in open air by comparison.

[MRAScott]

Probably not worth the effort then.  

If you're really worried that it's LLC causing your problem, try lowering the LLC and see if it lowers that voltage spike.  Lowering LLC may destabilize your overclock, though.  

 

 

What? No! Why would you say that? There's a specific entry in my thread saying why that's a bad idea and how it's unnecessary.  

You're right - I had a brain-fart.

 

I was thinking of the preparation of heatsinks - thermal/mechanical interfaces - not of electronics.

 

But for the afore-mentioned (heatsinks and similar), the way to go is indeed lighter-petrol followed by isopropyl.

 

Sorry for the mis-understanding.

[MRAScott]

There's no benefit to this at all.  isopropyl alcohol is a much more effective cleaning solvent, as butane will evaporate very quickly in open air by comparison.

 

There is, absolutely, a benefit in using lighter-fuel.

 

I speak from experience - lighter petrol disolves grease, and generally a far greater range of contaminants, much more quickly and easily than isopropyl.

 

The isopropyl can then be used to mop up the petroleum, and whatever other residue is left.

 

ETA >> Just noticed you mentioned "butane". I'm talking about petrol - the stuff you put in Zippo's

[DildorTheDecent]

3570K 4.6GHz 1.310V. LLC on Extreme makes it appear as 1.308V, in CPU-Z.

[MRAScott]

3570K 4.6GHz 1.310V. LLC on Extreme makes it appear as 1.308V, in CPU-Z.

I'm going to have to do a lot of video encoding in the near future.

 

I'll need an Ivy Bridge (8-thread) CPU.

 

My plan is to implement one of the following;

 

-  de-lid the Ivy, then re-attach the IHS to the die and PCB with JB-weld, and use a closed-loop water-cooler.

 

- de-lid the Ivy, then attach a closed-loop cooler directly to the die (no IHS), having attached a perfectly flat copper spacer to the cooler HS (probably also using JB-Weld).

 

By "closed loop" water-cooler, I mean the likes of the Corsair "hydro series", which reside inside the case.

[MRAScott]

3570K 4.6GHz 1.310V. LLC on Extreme makes it appear as 1.308V, in CPU-Z.

A whole 2 mV difference? Are you sure?

[sgloux3470]

There is, absolutely, a benefit in using lighter-fuel.

 

I speak from experience - lighter petrol disolves grease, and generally a far greater range of contaminants, much more quickly and easily than isopropyl.

 

The isopropyl can then be used to mop up the petroleum, and whatever other residue is left.

 

ETA >> Just noticed you mentioned "butane". I'm talking about petrol - the stuff you put in Zippo's

Lighters typically use butane as fuel.  Isopropyl alcohol works better as a solvent than butane, as butane will evaporate at -1C and thus cannot dissolve anything very well.

 

I sure as shit hope you're not putting gasoline in your zippo's and cleaning your computer with it.  

[DildorTheDecent]

A whole 2 mV difference? Are you sure?

Yes. My motherboard is the Z77X-UP7. If it can't hold voltages then nothing can.

Also your previous post about delid, I'd leave the IHS on. Wouldn't want to risk cracking the die.

[FireFox]

Derp