45nm Intel CPU Failure Resulting From Loadline Calibration Enabled

[nuclearjock]

In an article from Anandtech written about a year ago, the apparent negative aspects of overclocking with loadline calibration enabled were discussed.

In the past ten months I have built a total of twelve 45nm Asus/Intel based gaming systems. Eleven for friends and co-workers and one for myself. All twelve systems are water cooled, and run anywhere from 20-35% overclock 24/7. ALL systems have loadline calibration enabled and there have been zero cpu failures in any of these systems.

Discussions in another forum aluded to the fact that certain individuals here at XS have experienced failure or fried 45nm chips and feel LLC is to blame.

If you're one of those to be so unfortunate, I would appreciate a short reply with your thoughts and observatons.

TIA,

Nuke

[Dee]

Interesting. I have LLC enabled and I'd like more info on this too.

[Cooper]

now we need a full report on those dead cpus. clock speeds, voltage, cooling and the reasons they set to blame LLC - and that's not full report
something tells me it's absolutely irrelevant.

[ReverendMaynard]

Originally Posted by Cooper

now we need a full report on those dead cpus. clock speeds, voltage, cooling and the reasons they set to blame LLC - and that's not full report
something tells me it's absolutely irrelevant.

look at the date of that article. If this were true, allot of people would be killing cpus by now.

[Dee]

Originally Posted by ReverendMaynard

look at the date of that article. If this were true, allot of people would be killing cpus by now.

Maybe they did kill them... so now they have no Internet and can't reply to say they killed them!

[platojones]

Originally Posted by nuclearjock

In an article from Anandtech written about a year ago, the apparent negative aspects of overclocking with loadline calibration enabled were discussed.

In the past ten months I have built a total of twelve 45nm Asus/Intel based gaming systems. Eleven for friends and co-workers and one for myself. All twelve systems are water cooled, and run anywhere from 20-35% overclock 24/7. ALL systems have loadline calibration enabled and there have been zero cpu failures in any of these systems.

Discussions in another forum aluded to the fact that certain individuals here at XS have experienced failure or fried 45nm chips and feel LLC is to blame.

If you're one of those to be so unfortunate, I would appreciate a short reply with your thoughts and observatons.

TIA,

Nuke

This is not mysterious at all. LLC is a hack introduced by several MB makers to satisfy a market demand...based many clueless OCer's understanding of basic electrical engineering concepts..specifically, the concepts that are behind vdroop in the first place. With LLC enabled, transient voltage spikes when loading and unloading the CPU (games and benchmark programs, specifically) introduce spikes well outside the range of the limits tolerated by 45 nm chips. CPU degradation can be either a quick or slow process, depending on the base Vcore setting of the CPU (which is typically, much higher than VID on overclocked systems). Either way, LLC will rapidly reduce the life of a CPU by introducing massive transient overvolts every time the CPU voltages change in a dramatic way (gaming, for instance). All the MB makers did was to formally introduce a way to implement the infamous 'vdroop pencil mod' into the BIOS, requiring the clueless user to actually do it himself. So, the warning about LLC (and vdroop mods) has been there for a long time. If you burned your CPU using it...it is your fault. Caveat emptor.

[ThugsRook]

LLC is a lie to yourself.
it effectivly raises your voltage 0.1v but ppl feel safe cause they never actually see that. some ppl actually think its great that their voltage raises during load.

[ahmad]

I don't remember hearing reports of failure. I do recall several people noting instability with LLC enabled.

Personally, the effects of LLC scare me. I don't like seeing the load voltage go above what I set in the bios, to me that seems unreliable and I'd much rather set a higher setting in the bios. It is not like we don't have a higher voltage option.. it goes all the way to 2.1v!!

[philbrown23]

well this actually happened to my E8500 a while back.

[whe3ls]

i have LLC on a DS4 but i dont think its like Asus LLC

[platojones]

Originally Posted by whe3ls

i have LLC on a DS4 but i dont think its like Asus LLC

Well..it's your system. I would like to know, however, why you don't think it is like Asus LLC...the meaning of LLC is pretty clear...

[whe3ls]

well, when i had my max extreme when i went from idle to load my vcore didnt change. but with my gigabyte ds4 i still get i drop but its about half of what i get with it disabled

[dnottis]

Pencil modded for vdroop, disabled ever since, didn't need it. I did use it for along time with no issues.

[platojones]

Originally Posted by whe3ls

well, when i had my max extreme when i went from idle to load my vcore didnt change. but with my gigabyte ds4 i still get i drop but its about half of what i get with it disabled

Well...is it a safe vdroop? Did you read and understand the Anandtech article (or the Intel white papers on the subject?) Nobody is responsible for what volts you pump through your CPU but you when you OC...but I hope you realize that when you OC, you assume responsibility for the outcome...not Intel or the MB maker. It is up to you that the transient volts don't go above the maximum VID.

[platojones]

Originally Posted by dnottis

Pencil modded for vdroop, disabled ever since, didn't need it. I did use it for along time with no issues.

65nm chips are much more tolerant, voltage wise. 45nm chips burn easy and fast when exposed to overvolts...it's a new world, but some people don't seem to realize it yet.

[nuclearjock]

OK, let's say I have a Q9650 that's prime stable @ 1.33v with LLC enabled. SYStem idles and runs fully loaded @1.33v, no change.

I disable LLC, and find that I need ~1.38v unloaded to give me 1.33 under load which will achieve Prime stability.

Would you consider no load above 1.36v (Intel's recommended max) to be safer than enabling LLC and remaining at 1.33 constantly.

That is to say, are the voltage excursions (transient spikes) more damaging with LLC enabled than idling at 1.38v with LLC disabled??

[platojones]

Originally Posted by nuclearjock

OK, let's say I have a Q9650 that's prime stable @ 1.33v with LLC enabled. SYStem idles and runs fully loaded @1.33v, no change.

I disable LLC, and find that I need ~1.38v unloaded to give me 1.33 under load which will achieve Prime stability.

Would you consider no load above 1.36v (Intel's recommended max) to be safer than enabling LLC and remaining at 1.33 constantly.

That is to say, are the voltage excursions (transient spikes) more damaging with LLC enabled than idling at 1.38v with LLC disabled??

Yes..you are safer (not absolutely safe, since Intel says anything above VID for your chip is theoretically unsafe...a risk all OCer's take)...running at a higher no-load (but under the max VID for your chip) voltage is much better than allowing transients to occur above the highest rated voltage...any high transient volts can knock your chip out instantly at any time...and will certainly degrade it faster than the higher temps or higher unloaded volts incurred with a non-LLC system.

[adamsleath]

I disable LLC, and find that I need ~1.38v unloaded to give me 1.33 under load which will achieve Prime stability.

the point being that say hypothetically you use LLC at 1.33Vcore; will it "spike" above 1.38V?

my p5k has 0.05 vdroop.
my striker had 0.02 vdroop.

and most boards vary.

i think LLC would increase likelihood of damage when you have set your base vcore at a higher level.

eg. 1.38Vcore + LLC?
raises the point of highest safe voltage for a 45nm chip, and just exactly what vdroop buffer is recommended for these "spikes"

[ThugsRook]

Originally Posted by nuclearjock

Would you consider no load above 1.36v (Intel's recommended max) to be safer than enabling LLC and remaining at 1.33 constantly.

That is to say, are the voltage excursions (transient spikes) more damaging with LLC enabled than idling at 1.38v with LLC disabled??

spikes with LLC enabled happen above spec, spikes with LLC disabled happen below spec.

the entire point of the spec is to set a max, not the constant voltage. LLC ignores and removes all that, its outta spec 100% of the time.

[adamsleath]

spikes with LLC enabled happen above spec, spikes with LLC disabled happen below spec.

how do you know that? when running an oced chip above the recommended voltage it is always out of spec regardless of LLC or no LLC

if you set voltage below max spec with LLC enabled it may occasionally spike above the max spec...vs an oced chip that is set above spec and vdroops still above spec.

[branimir]

I have maximus formula. What was noticeable with bios 907, e8400 and
llc enabled is that everest fpu benchmarks rebooted windows with any
voltage, while prime95 was stable at same voltage. That means I needed higher
voltage to be stable with llc on (probably). But, with update to bios 1207
, seems that asus corrected that and now llc works ok on my system
(no reboots on everest fpu tests).

Greets.

[adamsleath]

it's interesting though, but to be careful of LLC particularly at 1.36 or higher volts...i suppose.

but what about crazy volts under phase? at 1.6V?

[platojones]

Originally Posted by adamsleath

the point being that say hypothetically you use LLC at 1.33Vcore; will it "spike" above 1.38V?

Well, read the article posted by the op. It has graphs and everything...it depends on the VID of the particular chip. It can happen to a 45 nm or 65 nm...and throw in the fact that each chip that is manufactured has it's own VID (each individual chip)...So you may have a golden chip that withstands the most obscene overvolt incursions...or a glass chip that dies the minute you go above the max VID for that chip. Definitly read the Intel spec sheets for your specific chip...if you are going above the highest VID voltages (or any volts for the other chips in your chipset), then your chances of a burned system go up dramatically...unless you got really lucky and found a really tolerant set of chips...here's a place to start...for the q9550:

http://download.intel.com/design/pro...hts/318726.pdf

It's all there for that chip. Read it and understand it before you OC, and especially before you try LLC. I see a lot of people posting pictures of their Q9450 or Q9550 at 4GHz with a Vcore of 1.4 and LLC enabled...well, if you look up their subsequent posts, you'll probably hear ing about how their POS Intel yorkie died a premature death...and, btw, they are going to RMA the thing back to newegg because it just totally sux...har har har.

Many people just get lucky (I think they are over-represented on these OC forums, BTW). But a lot of people here just want to get more bang for the buck and can't afford a new Chip or MB every other week. So, for those of you, like me, who need this thing to last for at least a year...pay attention to these limits....and stay away from LLC.

[ThugsRook]

Originally Posted by adamsleath

how do you know that? when running an oced chip above the recommended voltage it is always out of spec regardless of LLC or no LLC

if you set voltage below max spec with LLC enabled it may occasionally spike above the max spec...vs an oced chip that is set above spec and vdroops still above spec.

the assumption is that you are within spec to begin with. if you are overclocking beyond VID that is one thing, overclocking beyond VID with LLC enabled is quite another.

honestly, its just a way of fooling yourself into thinking you are running 0.1v less then you really are.

[platojones]

Originally Posted by ThugsRook

the assumption is that you are within spec to begin with. if you are overclocking beyond VID that is one thing, overclocking beyond VID with LLC enabled is quite another.

honestly, its just a way of fooling yourself into thinking you are running 0.1v less then you really are.

Exaclty...it's marketing...not engineering. It's just a dangerous short cut to allow people to think they have a safe, stable overclock with 15 minutes of effort instead of the many tedious hours (most of time, but not always) it takes to get a good, solid, relatively safe overclock...think about it...if it were that easy, why do they have the QX series CPUs.