Help me understand load line calibration.

[Maximilian]

Lets say you have an overclock of 4.5ghz and it needs 1.3v to be stable. It seems to me you can have it one of two ways:

Say the VID is 1.35v
LLC on -0.05 offset voltage, cpu-z reports 1.3v under load (1.35 - 0.05)

Say the VID is 1.35v and vdroop brings this is 1.25v under load
LLC off +0.05 offset voltage cpu-z reports 1.3v under load (1.25 + 0.05)

Is this correct?

Is there any actual difference between either method if the resulting vcore under load is the same?

 

[Idontcare]

Lets say you have an overclock of 4.5ghz and it needs 1.3v to be stable. It seems to me you can have it one of two ways:

Say the VID is 1.35v
LLC on -0.05 offset voltage, cpu-z reports 1.3v under load (1.35 - 0.05)

Say the VID is 1.35v and vdroop brings this is 1.25v under load
LLC off +0.05 offset voltage cpu-z reports 1.3v under load (1.25 + 0.05)

Is this correct?

Yes, correct.

Is there any actual difference between either method if the resulting vcore under load is the same?

In theory, yes there is a difference, in reality though there is no practical difference.

 

[Maximilian]

Yes, correct.




In theory, yes there is a difference, in reality though there is no practical difference.

Excellent, good to know.

Ill stick with LLC off then since its already like that anyways, saves me messing about more :thumbsup:

 

[Idontcare]

If you are using the offset method, be aware that it effects both the loaded and unloaded voltages.

In your example you went from a -0.05V offset to a +0.05V offset. At load that makes no difference in the voltage, in your example, but the idle voltage will be increased by +0.1V versus what it was when using LLC and the -0.05V.

 

[Maximilian]

If you are using the offset method, be aware that it effects both the loaded and unloaded voltages.

In your example you went from a -0.05V offset to a +0.05V offset. At load that makes no difference in the voltage, in your example, but the idle voltage will be increased by +0.1V versus what it was when using LLC and the -0.05V.

Yes, with the actual settings on my chip i went from -0.2v at 4.4ghz to +0.5v at 4.7ghz, LLC off both times. I originally used 4.4ghz because i thought it would be more efficient at idle due to the lower voltage however both times idle power was at 117w at the wall. Overall the increased idle voltage didn't seem to make much of an impact.

 

[Idontcare]

Yes, with the actual settings on my chip i went from -0.2v at 4.4ghz to +0.5v at 4.7ghz, LLC off both times. I originally used 4.4ghz because i thought it would be more efficient at idle due to the lower voltage however both times idle power was at 117w at the wall. Overall the increased idle voltage didn't seem to make much of an impact.

Yep, that's where the "in reality it doesn't make a practical difference" comes in

 

[cmdrdredd]

Supposedly LLC causes voltage to fluctuate more and software like cpu-z can't pick it up. and won't show the spikes up and down.

This still true?

 

[Idontcare]

Supposedly LLC causes voltage to fluctuate more and software like cpu-z can't pick it up. and won't show the spikes up and down.

This still true?

In theory, yes that is true.

Does it make a practical difference though? Not really.

As time passes it is looking more and more like the "LLC debate" is akin to the now irrelevant advice "don't start your car and immediately drive it on a cold day! you'll dextroy yer barings!"...advice that was valid and relevant but only if you happen to still be driving a car manufactured in 1963.

 

[C.C.]

Supposedly LLC causes voltage to fluctuate more and software like cpu-z can't pick it up. and won't show the spikes up and down.

This still true?

I am using an offset voltage with LLC @ 1 on my system in signature, and CPU-Z seems to report the voltage swings just fine?

 

[Grooveriding]

LLC seems more a hindrance than anything else on my system. It causes the VRMs on my motherboard to get hotter the higher it's set. I leave it on medium and adjust voltages from there. When I still used air cooling anything over medium LLC and my system would shut down from the VRMs overheating.

 

[cmdrdredd]

I am using an offset voltage with LLC @ 1 on my system in signature, and CPU-Z seems to report the voltage swings just fine?

What I mean is it was mentioned before something about llc having big spikes that don't get picked up because they happen too quickly. I never thought it to be an issue really. There are probably a couple old threads if I did a search for it. Thought someone might remember what I was referring to.

I had always assumed, and probably rightly so that manufacturers wouldn't put llc there if it could kill your CPU.

 

[Plimogz]

There are probably a couple old threads if I did a search for it. Thought someone might remember what I was referring to.

Oh, there certainly are. But seeing as how the best of them were both started and then championed by Idontcare, and he's posted thrice in this thread, the point is likely moot.

To me, LLC certainly seems like an obviously good deal at first, until I remember how freakin often per second my CPU may be switching power level from low to high and back to low again... Then I thank heavens that I never used the BIOS revisions for my motherboard which included 10 different LLC levels; that could've turned me off the whole tweaking thing altogether.

Now I'm just happy that changing the voltage offset setting in my BIOS greys out the LLC option.

 

[cmdrdredd]

I gave up using llc when I found my voltage could be a little lower (which reduced temp a bit) with it off. When I turn it on my voltage was always going to be a bit higher for some reason.

 

[2is]

Problem with LLC and load voltage is that CPU's need voltage even when they're not at full load, and that's where LLC can cause problems, particularly the higher levels of LLC.

All LLC is doing is boosting voltage under load situations, which causes people to see that their load voltage is higher than it really needs to be... Enter negative offset to correct that, at the expense of lowering voltage across the entire frequency/load range where LLC is NOT providing a boost.

Another potential issue is transient voltage spikes when the CPU goes from full load to idle.

Personally, for mild-moderate overclocks, I'd just rather use a +offset. It may give me slightly higher than necessary voltage at idle/low load, of which the consequences are nothing more than a negligible increase in idle power consumption, which is better than potential instability and voltage spikes going into my CPU.

 

[Idontcare]

...the consequences are nothing more than a negligible increase in idle power consumption, which is better than potential instability and voltage spikes going into my CPU.

If you are overclocking to any appreciable degree, with or without over-volting on top of it, then you are already allowing for a portfolio of consequences, and accepting the risk thereof, that far and away exceeds that of enabling LLC on your mobo.

Which is why overclocking expressly voids your warranty (not counting the Intel performance tuning warranty), whereas enabling LLC on your mobo does not

 

[2is]

If you are overclocking to any appreciable degree, with or without over-volting on top of it, then you are already allowing for a portfolio of consequences, and accepting the risk thereof, that far and away exceeds that of enabling LLC on your mobo.

Which is why overclocking expressly voids your warranty (not counting the Intel performance tuning warranty), whereas enabling LLC on your mobo does not

Nothing here negates anything I said. There are different levels of "risk" Overclocking by 100MHz isn't the same as overclocking by 1GHz+over-volting+LLC, even if they both void your warranty. I'm not really sure what point you're trying to make here. If it's that your warranty gets voided, I already know that.

 

[Idontcare]

Nothing here negates anything I said. There are different levels of "risk" Overclocking by 100MHz isn't the same as overclocking by 1GHz+over-volting+LLC, even if they both void your warranty. I'm not really sure what point you're trying to make here. If it's that your warranty gets voided, I already know that.

Per your words you claim to know "which is better than...", to know which is better would require an understanding of the design tradeoffs that are actually involved at the electrical engineering level.

Given that the engineers who make the mobos and cpus really do understand which is better, and using LLC does not invalidate your warranty whereas straight up over-volting with the offset option in the bios does, it would appear that at best you are second-guessing the engineers and concluding you know better in contradiction to them.

Or I misinterpretted you post, which is also a possibility.

 

[2is]

IDC, I know you're a knowledgeable guy, but I have to ask, do you even know what LLC is? Maybe I too am misinterpreting you post, but based on what it looks like you're saying, it's a question I have to ask.

vdroop is part of Intel's spec and LLC Is there to eliminate it, LLC is little more than a marketing term motherboard manufacturers use that essentially means over-volting under load. When you apply a load to a circuit voltage drops, without modifying the circuit itself, the only way to reduce the drop is by applying MORE voltage during that load state.

Weather it voids the warranty I don't know, if it doesn't, it only doesn't because of a loophole, not because it isn't over-volting, because it very much is over-volting the processor.

In conclusion:
It does over-volt the processor
It is against Intel's spec
The way it's used by many people, it can cause idle/low load voltage to be compromised
It's inherant nature can cause transient voltage spikes, much higher than raising your offset when the processor goes from full load to idle in an instant.

Now, I'd like to know why you think it's such a great idea.

 

[Idontcare]

From a purest standpoint I am not pro-LLC, I'm actually against it. I know what voltage spikes do to an IC. But in the non-academic realm, when we reduce this to the practical, LLC is just not the devil that laypeople like to make it out to be in forum settings like these. It is this over-attribution of fear and loathing for LLC that I find puzzling, so that is what I find myself following up on more so than the questions of how bad (or not) it is in reality.

That said, as far as I can determine, nowadays you technically can't overvolt an Intel or an AMD out of spec and invalidate the warranty because there are no voltage specifications for them in the first place. Period.

And if you have bought the latest AMD chip there is no way to technically overheat them and invalidate your warranty either, because a thermal spec doesn't exist for Vishera FX CPUs.

I was being facetious about the offset stuff in my comments way above. I realize my kind of humor is probably not mixing well with the text I'm typing here in this thread, I'm re-reading my posts now and am realizing they lack all the usual sarcastic emoticons I try to liberally pepper my posts with. So lord only knows just how offensive or argumentative you took my post(s), if you thought I was being a dick or something then I apologize, that wasn't at all what I had in my mind when I was drafting those responses to your posts. I have no beef with you whatsoever.

Not that you could, because the search here in these forums sucks horribly, but if you found my early posts on LLC I was staunchly against it. I have only changed my opinion recently (last 2-3 yrs) because the implementation have improved, they are not the crude implementations they were out there 5 yrs ago, they are more refined and do far less damage now (unless people do silly stuff like enabling the extreme setting in the BIOS ).

FWIW, to me the trade-offs you were referencing are pretty much half of one, six of the other.

At any rate, was just enjoying a bit of mental engagement is all. Its xmas eve and I've had a few, get to go "make a trip to the store" in an hour or so as my excuse for stepping out of the house so I can side-step into that garage and put on the santa suit for entertaining the kids with old "Santa's here!" routine

They get a visit from Santa before they go to bed on xmas eve, complete with a few of their presents as requested on their santa wish lists, and then Santa comes back and fills their stockings with goodies after they go to sleep. It is fun but I think this year they are going to figure out it is dad in that santa suit :hmm:

 

[2is]

Well... I'm not sure what half of one six of the other means exactly, I assume you mean equal tradeoffs though, but I don't agree... I'd say slighlty higher idle voltage than necessary is a lesser evil than lower than necessary with a spike on it's way there. That said, LLC or not, I hope you and your family have a Merry Christmas.

 

[Idontcare]

Well... I'm not sure what half of one six of the other means exactly, I assume you mean equal tradeoffs though, but I don't agree... I'd say slighlty higher idle voltage than necessary is a lesser evil than lower than necessary with a spike on it's way there. That said, LLC or not, I hope you and your family have a Merry Christmas.

Yes that is what the idiom means.

I don't get your math. You acknowledge having no idea on the specifics of the transient itself, neither its peak overshoot value nor its half-life decay rate, and yet you have confidently decided that it (even as an unknown) must be worst for your chip versus what you knowingly are doing to your chip by running it at a higher voltage during steady-state operation.

This is all duty-cycle driven, and if you understand the physical reason why voltage (of any kind) is bad for your chip then surely you would know you need more specific numbers to run through your calculations before you could ever have confidence in knowing which is better versus which is worse

This is what I meant above when I wrote that what I find baffling about LLC discussions is not whether people are right for the right reasons but that so many of these arguments appear to be right for the wrong reasons, and yet there is no shortage of confidence from the LLC naysayers despite their acceptance of the existence of all the unknowns.

FWIW, transient overshoot is part of Intel's spec. It is allowed and accounted for in Intel's expected device operation, within a specific overshoot and decay time of course, so simply claiming the existence of transient overshoot is not ample argumentation to justify the benefits of eliminating it as you appear to be attempting to do.

This is just an example Intel spec doc that readily came up in a google search on the topic. Page 14 has the relevant example information wherein the allowed Vcc_overshoot is 50mV with a max duration time of 25 micro-sec.

 

[2is]

You don't need to know exact figures. If you know what it's doing its pretty clear that LLC is the worse of two evils. The odd part is you actually agree and don't like using it but you're arguing for it anyway. Sounds like you're still bored and just want to be argumentative. That's cool. Ill continue to not like it and argue against it while you can continue to not like it and argue for it.