45nm Intel CPU Failure Resulting From Loadline Calibration Enabled

[dinos22]

i found my old post when raja over at OCX posted about this article from Kris

i found it a good read of course as you can see my comments in that thread however after having read it a few times i could not get my head around the idea of not using vdroop

Raja and Kris work together. IIRC Kris was looking at the thread when i posted the questions and chose not to reply
i dont know his reasons but i do know that it made no sense back then nor does it now even for extreme examples i posted about there.

I find vdroop is really required in xtreme cases where you use a lot of volts....for 24/7 machines i honestly dont think its really important to squeeze out everything out of the system (at least not for me as i dont do much aside from surfing and some work and also bit of photo/video editing)....so running at low volts and quieter system is my preference so even if i dont use LLC it doesnt matter but for LN2 benching its a must!

http://ocxtreme.org/forumenus/showthread.php?t=2208

[platojones]

right so please explain to me why a guy would step in here that has built a dozen machines past year say that he's used LLC on all his builds without a single issue. This is an extremely small example of the enormity of usage of LLC on this forum alone right. Can you please teach us more about the effects of LLC and what will happen to our systems then and how we would be able to properly determine/test the effects because we are all trying to learn something new every day. What would benchzoner have to do with his 5xrigs that have been hammered now for 2 months straight with different setups including high VTT, high VCORE, high PLL waaaaaaaaaaay beyond Intel specifications and what any user on XS dares use for 24/7 and some even for benching machines, without side effects of any kind........what do you suggest he does different.

we would love to learn something new from you if you actually have something to contribute. Afterall this is what xtremesystems is about>>> contributing right

Well, I'll be happy to take these one at a time, if you don't mind (and I don't really care whether you do or don't, so that was just a formality)

1) "right so please explain to me why a guy would step in here that has built a dozen machines past year say that he's used LLC on all his builds without a single issue."

I don't have to explain that or any other human behavior...he's claiming to run these chips so far out of documented specs, with no ill effects whatsoever for months...it's up to him to prove he's actually done it...he seems really shy about providing any evidence at all and, in fact, doesn't post that many updates...but then again, I was never sure what he was trying to prove...how long can he go with insane voltages? Funny as well, that these are chips that he has by the dozens...either very rich or someone (Intel or an MB manufacturer may be providing)...just wants to see if they will burn...not the standard I think many people are looking for. In fact, you may be surprised that some people may want to get more than 3 months of use out of their OC'ed rig before it degrades to the point of uselesness...as somebody once pointed out...this is "Xtreme Systems, not Xtremely Stable Systems"...but I will add that it is also not "Xtreme-ly Unstable Systems" either.

2) "This is an extremely small example of the enormity of usage of LLC on this forum alone right." Besides the fact that that this statement is 100% anectdotal...where are your numbers to prove it?

3) "Can you please teach us more about the effects of LLC and what will happen to our systems then and how we would be able to properly determine/test the effects because we are all trying to learn something new every day."

Actually read this article AnandTechs article and understand it..and Inte's white papers on this very subject...

4) Can you please teach us more about the effects of LLC and what will happen to our systems then and how we would be able to properly determine/test the effects because we are all trying to learn something new every day. What would benchzoner have to do with his 5xrigs that have been hammered now for 2 months straight with different setups including high VTT, high VCORE, high PLL waaaaaaaaaaay beyond Intel specifications and what any user on XS dares use for 24/7 and some even for benching machines, without side effects of any kind........what do you suggest he does different.

I can ignore this because it's explained above...how many years did this kind of rig run? What were the actual voltages? Exactly what does he mean when he says it did not degrade? Did he run 30 cycles of linpack? NO! He didn't. Systems usually only show the first signs of degradation under stress...then later degrade to the point of being useless...sometimes fast and sometimes slow.

I will say one more thing...degredation is not an easily observable effect...until it becomes observable...then it is too late. This is all based on actual physical laws...not what you think you see or don't see. Physics doesn't swing that way. All I have ever seen here is is very weak anectdotal claims that nothing bad has happened the the chips...but no follow-up stress tests....or actual proof.

Like I said, claiming that, based on a few months of extreme voltages, LLC enabled, etc, doesn't prove a thing...it proves that Intel over-engineered their chips pretty well to withstand more than they claimed...but Intel also says that you are degrading your chips much faster than you know.

All I am asking you to do is to admit that...this is not safe, you enjoy it, but for somebody who may come here asking advice about a stable OC, that will last them some reasonable period of time (assuming they don't have the funds to by a new chip every 90 days), the get something that approximates a correct answer...

That is my contribution here and if you don't like...I'm ok with that.

[platojones]

i found it a good read of course as you can see my comments in that thread however after having read it a few times i could not get my head around the idea of not using vdroop

How is that their problem? You don't comprehend it...so what.

Raja and Kris work together. IIRC Kris was looking at the thread when i posted the questions and chose not to reply
i dont know his reasons but i do know that it made no sense back then nor does it now even for extreme examples i posted about there.

Again, so what...they saw a lot of posts like that on the thread...all wrong...why should they single your incorrect assumptions out and explain very fundamental facts just for you, that they just, very eloquently, explained in the article. Unless you are close personal friends with these guys...which maybe isn't the case.

I find vdroop is really required in xtreme cases where you use a lot of volts....for 24/7 machines i honestly dont think its really important to squeeze out everything out of the system (at least not for me as i dont do much aside from surfing and some work and also bit of photo/video editing)....so running at low volts and quieter system is my preference so even if i dont use LLC it doesnt matter but for LN2 benching its a must!

OK, good personal choice for you...ignores the physics and on some level it makes you personally happy...doesn't negate any point I've made.

http://ocxtreme.org/forumenus/showthread.php?t=2208

No need for me to even look...I trust you...I actually believe you when you say you did post these things on AnandTech and that you were ignored...there, I'm not so bad after all, am I? Just an a friendly old crumudgeon.

[dinos22]

OK, good personal choice for you...ignores the physics and on some level it makes you personally happy...doesn't negate any point I've made.

i have to comment on the vdroop suggestion to leave it disabled. When you are pushing your overclocking with LN2 you need this enabled because of a very simple reason. if you don't leave it enabled you will be selecting extra 0.1v or more to reach the same stable overclock. For example, if you are pushing your QX9650 to 5.6GHz for 3D and you are using 1.95v with load calibration you will received that sort of value in windows under load. To do the same without the option you would need to select 2.05v-2.1v to get the sort of volts to get the CPU to finish those runs with enough vcore. Is Kris suggesting to people to disable loadline calibration and stick 2.1v into the chip via bios. We know that vdrop value will make that probably around 2v or maybe a touch more in idle but that will still be higher than loadline calibrated idle/load with only a microsecond spike at the start of the benchmark. Another thing is does chris use prime to explain vcore oscillations? How many overclockers with SS/DICE/Cascade/LN2 load the chip that hard, even every days systems. I don't think many do at all. Most people are usually just benching Futuremark or single thread of superpi which as a result usually doesn't load the CPU as hard as those graphs actually indicate. Maybe he had air or watercooling users in mind

so what would you do then if you had the choice while benching on LN2
would you still disable droop?

[platojones]

so what would you do then if you had the choice while benching on LN2
would you still disable droop?

I would hope you could guess the answer to that by now...and, BTW, what makes benching LN2 so much more important than benching anything else...are you are professional bencher? I'm providing a different point of view here...if there is a clear, compelling choice between hardware stability (and relative longevity) plus system stabililty and raw overclocks at any price...I always pick the former, no matter what....even if *gasp* I have to underclock to achieve it if the CPU/MB/Memory is crappy enough.

[mrcape]

Yo Dinos, I always left it disabled on the LP UT P35, including benching with LN. I had a hell of a time breaking 600fsb with that board, maybe that's what was holding me back!

I think the argument whether or not to use the feature is only applicable within certain user profiles trying to attain the same goals.

[dinos22]

humour me please

let's pretend that you are a shooting for W.R.s and you need those last few MHz out of the CPU and only way to reach them is to run your CPU at certain vcore. You are risking to kill a really nice CPU and have to pick lesser of two evils so to speak

what would you pick

background to the story is

When you are pushing your overclocking with LN2 you need this enabled because of a very simple reason. if you don't leave it enabled you will be selecting extra 0.1v or more to reach the same stable overclock. For example, if you are pushing your QX9650 to 5.6GHz for 3D and you are using 1.95v with load calibration you will received that sort of value in windows under load. To do the same without the option you would need to select 2.05v-2.1v to get the sort of volts to get the CPU to finish those runs with enough vcore. Is Kris suggesting to people to disable loadline calibration and stick 2.1v into the chip via bios. We know that vdrop value will make that probably around 2v or maybe a touch more in idle but that will still be higher than loadline calibrated idle/load with only a microsecond spike at the start of the benchmark

A) vdroop disabled

B) vdroop enabled

Yo Dinos, I always left it disabled on the LP UT P35, including benching with LN. I had a hell of a time breaking 600fsb with that board, maybe that's what was holding me back!

I think the argument whether or not to use the feature is only applicable within certain user profiles trying to attain the same goals.

i dont think vdroop has anything to do with FSB clocks

[platojones]

Let me just say one more time too...I'm not against anybody using LLC at all. Just understand what it is actually doing and don't make inaccurate statements about it...don't unintentionally mislead others...if you don't know what you are talking about, then say it up front...if new OCers know the facts and still want to use it...no problems. But to claim as a matter of fact that it doesn't contain additional risk to your chip is not true...so just don't claim that, no matter how many annecdotal stories you've heard that personally convince you that there is no significant risk...the physics of CPUs say you are wrong...that's all I am pleading for here.

[platojones]

humour me please

let's pretend that you are a shooting for W.R.s and you need those last few MHz out of the CPU and only way to reach them is to run your CPU at certain vcore. You are risking to kill a really nice CPU and have to pick lesser of two evils so to speak

what would you pick

background to the story is


A) vdroop disabled

B) vdroop enabled



i dont think vdroop has anything to do with FSB clocks

I can only ask you...what is the point of the WR? What are the rules...for system configuration, for proving stability...it all depends on the standards. A WR with no rules is meaningless. So if the WR rules say LLC is allowed, then sure...but for real people trying to squeeze an inch on their processing power, but who need a machine that is rock solid...who cares about the the WR anway? I guess I would have to say, and I may be the only one here...I'm not a professional OCer with 50 chips and MB's in my closet...I upgrade systems about every 2 years...I'm a professional software engineer, so I need a system I can trust. I just happen to like the idea that my system is as well tuned and fast as it can be...so I may be a fish out of water in the forums...and I suspect I am. Oh, well, maybe there is some room for me here...maybe not.

[mrcape]

B) vdroop enabled



i dont think vdroop has anything to do with FSB clocks

Tx, think I'm confusing/reversing the effect of LLC.

[nuclearjock]

Well, I'll be happy to take these one at a time, if you don't mind (and I don't really care whether you do or don't, so that was just a formality)

1) "right so please explain to me why a guy would step in here that has built a dozen machines past year say that he's used LLC on all his builds without a single issue."

I don't have to explain that or any other human behavior...he's claiming to run these chips so far out of documented specs, with no ill effects whatsoever for months...it's up to him to prove he's actually done it...he seems really shy about providing any evidence at all and, in fact, doesn't post that many updates...but then again, I was never sure what he was trying to prove...how long can he go with insane voltages?

I started this thread and I'm not sure if you're referring to me, but my builds were for friends who enjoy gaming as do I.

I've never had a problem with LLC enabled so I set up their rigs with LLC enabled as well.

Then I heard someone ( in the "other forum") going on about how LLC can be detrimental, I read the Anandtech article and was left somewhat confused.

I'm simply wondering that in this vast group of users who overclock in a substantial way, if there was clear observation that LLC was a bad thing to enable.

If you're referring to BZ, then I'll maybe he can respond.

Anyway, LLC is still enabled on all the rigs I've built that have the option available and everything seems to be ok.

[carepolice]

If i set 1.312v in the bios my voltage in windows reports 1.288v on load it goes to 1.296v and stays there.
.

Works like it should

[003]

Seems like a lot of ing over nothing is going on here. Fact is, LLC does not damage 45nm CPUs. I don't care if you have your degree in electrical engineering and fancy graphs to back up claims that LLC will damage 45nm CPUs. Because the fact still stands that it has not damaged them. Mine, nor anybody elses. How do your fancy graphs and degrees explain that?

[Leeghoofd]

So are we still on this... seems things are heating up...

Look some boards seem to be picky with or without LLC... doing damage to the CPU with LLC enabled, it could happen if you wank high volts through it, but there's more to doing damage to a CPU then LLC alone...I read a post by Gary from Anandtech now has almost 10 CPU's going down the drain, why was this never published ?...

Bad PSU's can also be the cause of voltage overshoots... too high PLL, FSB Term, bad cooling,...,ram capping out can also be to blaim an unstable OC... you are the tweaker you need to find the sweetspot for your hardware...

I have absolutely no degree in electrical science... and it seems to me people are obsessed with charts, graphs and white sheets...

Some rules of thumb from the old skool crowd :

If you want to OC safe with max longetivity: OC max 20%... (but everyone wants 500FSB now)
Second rule :please always use a digital volt meter when OC'ing... volts in bios might NOT correspond to what you actually are pumping to the power circuitry of ya precious...
Thirdly : Take it slow, I'm still OC'ing with a burn in at start, then slowly going up a few hundred Mhz at a time...
Fourthly : make sure all ya hardware/cooling is up to it... (I even saw people running OC's with the PSU upside down, so it got almost zero ventilation)...
Last but not least : do ya homework first : reading through wicked forums and make up ya mind if you want to violate white sheets or not... and it's not because ya neighbour can do them clocks that your similar rig can do them aswell...

I've been commented on giving insane volts for my CPU's to users... because I use them, you are NOT obliged to apply them... but they can be a start and might help you to post at the required settings... fine tweaking is up to you... each hardware combo is different... try to get the lowest possible voltage... and rememer when you OC things can go wrong and permanent damage can occur... you are the one that has to take responsability for that, as you are running hardware out of spec... not the guy that gave you some settings because he has got 5000 posts or more. My four 45nm CPU's are all still running strong, besides the one I messed up myself ( one of the reasons I got a DMM now ) but I learned it the hard way and I will not make that mistake again... maybe an I7 soon down the drain lol...

And for your info Benchzowner and Dinos are very highly respected in the Oc'ing community...( if they looked pretty they would be sexy too) They don't need this kind of attitude, as they have more hardware to mess with then we all ever could get together... Due to their contributions and efforts, hardware/bios changes are made in our benefit...
If you want safe clocks be my guest, but if you want more than the average Joe it's time to take risks... this forum has evolved big time from Die hard, voltage modifying etc overclocking to a more general OC forum where everybody can join , share feelings etc... if you wank 1.5 Vcore through your 45nm, cpu who did the input in the bios ? Me or you ? if you don't feel comfy with it, don't apply it plz...

[Utnorris]

Personally, the problem I have with the Anand article and everyone that was claiming LLC was damaging 45nm chips is that not one reproduced the results. To me, it's all theory until you can reproduce the results. I spoke with an Asus tech that does LN2 and attends overclocking contests and while he may not be an expert, he basically told me that LLC ads or subtracts voltage to maintain a steady vcore. I cannot see how this would be bad since it happens on the ms level. If this type of adjustment happening that quickly can kill a chip than so can noise on your psu lines. Until someone can reproduce the effect and show where they were able to kill their chip with LLC I am going to trust the manufacture that made the board. Personally, I have found that my system is more stable at a lower voltage with LLC enabled and therefore I will continue to use it. Lastly, most of the discussion on LLC has come about based on what one person states in one forum and then it gets repeated in another and before too long something that was just a theory or misstated becomes fact. Again, reproduce the results and I will believe them.

[Zucker2k]

So XS overclockers are smarter than Intel engineers Intel specifies max vcore and such too, yet we go way above that and suffer no consequence, does that mean too much vcore isn't a bad thing?

[SergioRoadster]

Happens way too fast to register on CPU-Z...it is jumping every time you load or unload the CPU, but for a very short period of time.

Thanks for clearing that out. Then must be sort of a micro-second thing.

[Utnorris]

So XS overclockers are smarter than Intel engineers Intel specifies max vcore and such too, yet we go way above that and suffer no consequence, does that mean too much vcore isn't a bad thing?

I do not think anyone is claiming to be smarter than an Intel Engineer, but since Intel has not commented on LLC, I really do not see your point. I get that we are throwing the chip out of spec, kinda the point, but the question that the OP asked was whether or not some anyone had experienced the same situation that the Anand article talked about, not whether or not too much vcore is bad for your chip. If you have experience with LLC, either good or bad, please do share.

[dinos22]

So are we still on this... seems things are heating up...

Look some boards seem to be picky with or without LLC... doing damage to the CPU with LLC enabled, it could happen if you wank high volts through it, but there's more to doing damage to a CPU then LLC alone...I read a post by Gary from Anandtech now has almost 10 CPU's going down the drain, why was this never published ?...
.

lol i think you may be referring to VTT here
if you can find a link where Gary says he killed 10xCPUs with LLC i'll seriously start considering him as a bit of a magician

[dinos22]

So XS overclockers are smarter than Intel engineers Intel specifies max vcore and such too, yet we go way above that and suffer no consequence, does that mean too much vcore isn't a bad thing?

Clearly none of us are even remotely close to having any intelligence when compared with Intel/AMD engineers
we are simply ignoring what they say and have done for years with AMD and Intel without the consequences they stated which leads me to believe that they understate their figures hard

[Leeghoofd]

lol i think you may be referring to VTT here
if you can find a link where Gary says he killed 10xCPUs with LLC i'll seriously start considering him as a bit of a magician

Well he has been vague what how and why... I know the Qx death was confirmed by Intel as VTT over load... the discussion here is that LLC is assumed as to be the culprit... I wanted to make clear ( in my bad english manner) that there's more to doing harm to a CPU then just LLC...

My quote :

but there's more to doing damage to a CPU then LLC alone...I read a post by Gary from Anandtech now has almost 10 CPU's going down the drain, why was this never published ?...

I find it weird they never let this loose on the big public... Did Intel forbid them to ? first thing I hear about a tester ruining so many chips in such a short time... Bingo13 is very vague here, not telling us anything : how, which clocks, cooling used, voltages... think it's time to call Mulder and Scully on this one

This is Gary's Quote mate in the degredation thread :

We tested 10 processors on three different boards, 3 died within five days, 2 showed serious degradation after 15 days (to the point they are no longer stable at stock settings), 1 showed light degradation after 25 days, four are still running fine and will under go degradation testing shortly. The failures were confirmed by Intel so there is no speculation as to the failure cause. We had six processors that developed problems and four that have not to date. Also, our test regimen stresses the platform in a variety of ways under varying conditions, not a single test running in a controlled loop. Whether that plays into the failures is something we are still investigating along with the chipset and board components.


Most of us run these CPU's out of specs (volt wise) and most CPU's can handle them daily punishments, sadly some give in sooner than others...but what I find weird is that some users have respected the white sheets and still have damaged their precious belongings... what was it ? bad PSU, bad CPU,...

to me it's the 45nm secrecy... I have 45nm CPU 's that act erratic and mobo chips too that seem have their own mind...

[dinos22]

yes Gary is talking about VTT there

[V3NOM]

so, what's the final verdict?

that LLC will not fry a CPU? that microsecond stuttering is not dangerous? this thread just seems relaly jumbled and confusing... i would just like a clear answer on whether LLC is dangerous or not.

thanks

[Falkentyne]

Check some of the other threads.
It's already been shown that it is MORE dangerous to raise idle voltage to compensate for vdroop's dips, than to have loadline/vdroop mod at a lower constant idle/load voltage), and have spikes to what your idle would be at the higher voltage (without vdroop). Pretty obvious actually.
1.35 idle/1.35 load with spikes up to 1.42v when switching states is safer than 1.42 idle/1.35 load, since the chip is exposed to less of the higher voltages. (Note that the average computer will spend more time idle than load, unless you're a cruncher or power off the computer as soon as you're done with any game or other load activity).

This is board dependent though. Proper working boards should have LLC no higher than bios setting or maybe .01v higher. If a board overvolts past bios setting at load, with LLC, in some cases, a lot more past cpu-z readings (taking with a DMM) then that's not normal and that's very bad.

[Rattle]

any thoughts, opinions or experiences with LLC on i7's ???

i really havnt been using it but I am creepiong up slowly and finding max OC at stock volts and having fun with it :p