45nm Intel CPU Failure Resulting From Loadline Calibration Enabled

[dnottis]

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.

yea, I've had a few 8400s and an E0 E8500 as well as the quad in my sig though.

[platojones]

yea, I've had a few 8400s and an E0 E8500 as well as the quad in my sig though.

Yeah, but how long did they run (if you constantly rotate chips, it's not quite the same is it?)..what was the vcore..what was the OC?...I assume you are saying you had LLC turned on as well, no?

[platojones]

Yeah, but how long did they run (if you constantly rotate chips, it's not quite the same is it?)..what was the vcore..what was the OC?...I assume you are saying you had LLC turned on as well, no?

BTW, this is a bit of an apples and oranges comparison...the quad 45nm are much harder to OC safely, I think. They require higher OC volts than than their dual core cousins anyway, putting the LLC risk threshold much lower. The Q6XXX series have the same risks...but, because the max VIDs are higher for 65nm chips, are more likely to allow higher OCs (assuming LLC is enabled) without degredation.

[adamsleath]

yadayada
ive actually tried a pencil mod on the p5k. and i cleaned it off because i wasnt having much luck with the OC
anywho...

[nugzo]

My current experiences with and without LLC are as follows... (Striker 2 Formula)

LLC enabled.... 1.34 bios, 1.36 windows idle and load. I need 1.36v for stability at 4400mhz. Runs great for now with no issues at all.

LLC disabled.... 1.3875 bios for 1.41v windows idle. 1.38v loaded. Blue screens within an hour of small FFT. Normally within 15 minutes.

I had same behavior with E8500 and E8600.

What could be causing this? For me i cant seem to get stabilty with LLC disabled even though my loaded voltage meets or exceeds the volts needed for prior stability with LLC enabled. This CPU is also stable at 1.46v @4500mhz for 19hrs Small FFT... Temps and voltages are a little high so i run at 4400 24/7. What would You do?

[adamsleath]

mm that makes me think i may have had vcore set too high...with the pencil mod..i dont remember experimenting much with the pencil shading on.

[TheScavenger]

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

HAHAHA!

In their frustration they swore to never use a computer or the internet again

[003]

LLC does not damage 45nm CPUs. I have been running my Q9450 at 1.425v with LLC enabled for many months now and I recently tested it with IntelBurnTest to make sure everything was still fine. Stable as a rock. I have gone so far as using 1.55v with LLC for benching sessions. And I have to significantly increase my voltage in order to remain stable without LLC.

[Dee]

HAHAHA!

In their frustration they swore to never use a computer or the internet again

Yes, and I can sympathise. I have come sooo close to saying "F**k it! This just isn't worth the headache! I hate computers!"


Anyway... I'm really not sure about LLC now. I have horrible vdroop without it but it kinda scares me. To disable it off or not... I don't know anymore.

[rge]

The Andandtech article only tested one mobo, different mobo have different vrm circuitry, and people with different mobo's tend to have widely different experiences with loadline...

Negative view:
for example anandtech reported with p5e3 and quad... "Imagine our confusion as we desperately struggle to understand why our system is Prime95 stable for days yet continues to crash under absolutely no load. What's more, in spite of the absence of droop and for reasons unknown, enabling this feature artificially raises our CPU's minimum stable core voltage at 4.0GHz from 1.28V to about 1.33V. As a result, our system uses more power under load than is otherwise necessary."

And another person with Asus P5k and E8400 reported similar problems with loadline "Apparently My original problem was related to Asus's Load-Line Calibration feature. It is sepposed to reduce vdrop and allow you to lower your voltages, but it was causing me nothing but problems..
It was actually giving me a higher vcore at load then idle?? it was bumping up the voltages as i upped the fsb??. Also it was giving me odd stability problems. I could run games and cpu intensive apps all day ,, but when exiting the programs it would crash almost every time? After disabeling this feature my system is running much more stable ... albeit with more voltage"
http://www.ocforums.com/showpost.php...32&postcount=4

Neutral view:
While loadline did not work so well on those setups, in my situation it seemed to be a wash or neutral. On my GB P35dq6 and E8400 for 4ghz 15hr prime stable,
For loadline on, 1.36 bios, 1.33 cpuz idle, 1.33 cpuz load (actual vdroop is ~.006 too small to be measured by mobo sensor with steps of .016v). I would assume my overshoot spike goes near 1.36 as bios setting is supposedly near max seen.
For loadline off, I need 1.33 bios, 1.33 cpuz idle, 1.33 cpuz load, and vdroop is ~.003 if GB is correct in that is may be half of vdroop. So maybe instead of spiking to 1.36, I spike to ?1.363. So for me it is a wash....so dont bother with it. I have run over a month both ways, and rock solid either way.

Positive view:
Many with positive results including nugzo from few posts ago.

People with different setups and especially with quads with larger vdroop seem to benefit sometimes, and that is where I would be interested in seeing more data. One mobo on Andandtech with some artist renditions of waveforms while informative and interesting, did not answer for me the question whether loadline was helpful or not. I would like to see multiple actual waveforms at different voltages on multiple motherboards and cpus. That may not be financially reasonable to do, and even then that would only answer part of the question. It would portray the true spike, idle, and load differences between loadline off vs on for different mobo's and cpu's. But it would not answer (in those instances where such occurs with loadline) whether a lower constant idle or load voltage but higher spikes versus higher constant idle or load voltage with lower spikes is more detrimental to cpu life.

[genec57]

OP asked for specific cases of chips running LLC that were fried. So far, there has not been one. There have been lots of pronouncements of the dangers of destroying chips due to the use of LLC.
I have seen many threads on this and other boards with dire warnings of the dangers of overclocking. The reports of actual chips destroyed or damaged as a result of overclocking or use of LLC have been very scarce (nonexistent?) indeed.
I am sure that at a certain point you will fry a chip but where is that point? I have never not ( how's that for a double negative?) used LLC; with no issues. I do pay attention to temperature and vCore within self-determined limits.
Really, if LLC and high overclock is so dire there should be lots of examples for us to learn from. I don't really care about engineering theory. I do care about real world examples, let's see them.

[BenchZowner]

Grabs the Mythbusting stick and posts:

First we had the High PLL, or high VTT ( FSB Termination ) killed our chips and we're sure and blah blah...
I also remember the "deadly" voltages in a table in the review of the Asus Striker II Extreme.
And then... a lunatic decides to test lots of CPUs under crazy conditions to see if they're degrading or even dieing when using high voltages ( and sometimes... insanely high voltages )... bang
And then... Core 2 Goes To Hollywood... 1.6Vcore LoadLine Calibration Enabled 2.0Vtt ( FSB Termination ) 2.8Vpll running F@H 24/7 @ Full load... lives on for days and days without a simple sign of degradation, nor death.

As for LLC itself... I've been running all my systems and CPUs with no Vdroop ( either Vmod or LLC enabled where applicable ) since the first Conroe E.S. appeared... that's more than 2 full years, with not a single dead CPU from LLC or high voltages... and everybody on these forums knows what kind of voltages I run no matter what cooling I'm using

[ST4]

imho, the source of the issue isn't llc, it's in the supply voltage itself. Case in point, why is there so much transient overshooting, undershooting, and ringing? Insufficient decoupling to clean out this noise, or bad grounding schemes? llc and the like is just masking these systemic problems, and Intel should tighten pk-pk ripple spec on the mobo vendors, but unfortunately they can't enforce.

[slim142]

I think I experienced this (and I think this was also the responsable for my video problems)

First, I would be 24 hours Prime95 stable, but then I would go watch a movie and all you know after 10 secs the PC freezes.
Then I would also leave the PC in idle, leave for 30 mins come back, move the mouse and 5 secs, restart.

I updated my BIOS and RMAed the board and problems were gone. Since then I still havent overclock my system (wating for water cooling), but once I do I hope LLC doesnt cause any problems.

[nuclearjock]

Well like I mentioned when opening the thread, 12 systems, no problems.

Went so far as to run the Intel Burn test on three of them today (including mine) and all's totally good.

I thought I would encounter a list of case horrors, but in actuality only a scary article which from an engineering point makes sense.

But I can't imagine mobo manufactures would release products with a feature as this that would be so lethal as to predictably consume expensive cpu's when enabled.

Let's face it, thery're having enough trouble with simple quality control let alone suicide bios options...

I'm leaving all mine turned on, I'll be the first to post a big note if any fail before i7. I've already been given the go ahead to upgrade six of these systems when Nehalem is on the street and with the meeger margin I charge for assembly, it just may be enough to build one for myself but that's another story.

[Dee]

I think I experienced this (and I think this was also the responsable for my video problems)

First, I would be 24 hours Prime95 stable, but then I would go watch a movie and all you know after 10 secs the PC freezes.
Then I would also leave the PC in idle, leave for 30 mins come back, move the mouse and 5 secs, restart.

I updated my BIOS and RMAed the board and problems were gone. Since then I still havent overclock my system (wating for water cooling), but once I do I hope LLC doesnt cause any problems.

Exact same problems as a system I recently built for a friend based on a Q6600 and an Asus P5K-E Wifi. We raised the CPU voltage from Auto (he doesn't OC) to 1.3v and the problems stopped except for one random reboot for no apparent reason.

[Ket]

I've run LLC enabled and disabled on my 65nm, but like anything common sense needs to be used. For me with LLC enabled I can set 1.5v in BIOS and its actually 1.48v idle, and about 1.47v load. With LLC disabled and using 1.5v my idle voltage is about 1.45v and loaded around 1.43\4v. I don't see LLC as "dangerously spiking", just adding enough to help maintain a more stable voltage.

[slim142]

Well so who do we blame? or who do we make responsable?

I hate to pay premium for LLC if its not going to help me at all, but oh well mileage may vary.

[adamsleath]

I would like to see multiple actual waveforms at different voltages on multiple motherboards and cpus

every motherboard will be different, even ones with the exact same model number will vary from board to board, or so ive noticed on various people's reports on vdroop and motherboards.
i bet some motherboards will not spike as much as others based on their componentry, but that is a guess.
...does it all come down to the v reg circuitry??
also will depend on the cpu, some are golden and others are average.

[mikeyakame]

If you set vTT higher than vCore then LLC will never damage a CPU. vTT acts as a transient overshoot buffer to protect the CPU, especially during power on when it spikes the highest. Problem is everybody wants to run low vTT without actually knowing what the hell it's designed to do. Even ANANDTECH I would have thought would have known this when they made these ridiculous claims. 45Nm chips are more delicate when it comes to instanteous overshoot, if you are running a good PSU that has excellent voltage output then you should never fry a CPU up this way. A poor PSU and low vTT could easily result in damage to a 45Nm chip at higher than 1.45V vCore.

[Dee]

So if my vCore is set to 1.4250 in BIOS (which gives 1.408 heavy load in Windows, according to CPU-Z) where should I set vTT, Mikey?

[mikeyakame]

Just slightly above what the real vCore is, if you get 1.41v idle/load by sensors, aim for around 10mV-20mV (0.01 - 0.02v) above the actual vCore. I know the P45 boards don;'t have vTT monitoring which makes determining it a little harder but if you are using LLC I would set the vTT to around 1.38 - 1.40v max and assume a slight overvoltage due to LLC which is safe to say. Otherwise if no LLC enabled set it around 1.44v. The Rampage Formula overshoots by about 40-50mV with LLC enabled. It's safer to assume this is true when you can't monitor it via sensors. vPLL would probably want to be around 1.60-1.64V. vPLL has a very easy method to gauge whether it's right or not with Prime95. All you need to do is monitor the difference between both threads for either core and how long apart they take to complete each loop. Dual core makes it a little tougher since OS related usage can effect the accuracy, but close what you don't need running while Priming and you shouldn't get too much error in ur results from outside system resource use.

on my 65nm Q6600 G0 I use 1.4375v vCore in bios which is 1.41-1.42V real with LLC enabled under load. vTT is set to 1.42V in bios which is 1.46V real, and vPLL is set to 1.62V in bios which is 1.68V real on my board. GTLREF is set to 0.65x for CPU, and 0.63x for NB. vNB is set to 1.47v in bios which is roughly the same real. Heres a quick Everest sensor paste of mine with my voltages. Southbridge PLL = SB Sata voltage in case you don't know. CPU is running on 8x multi and 3.45GHz, FSB at 430MHz, RAM at DDR2-1149 5-5-5-15 ( 4 x 1gb sticks PSC chips I believe) and PL 7 with Phase Pull ins on A1, A3, B1, B3 to PL 6. DRAM Static Read enabled, AI Clock twister at Stronger. DRAM Voltage at 2.26 or 2.28V in bios, and DRAM Controller REF +20mV

Motherboard Name Asus Maximus Extreme / Maximus Formula / Rampage Formula

Temperatures
Motherboard 28 °C (82 °F)
CPU 21 °C (70 °F)
CPU #1 / Core #1 31 °C (88 °F)
CPU #1 / Core #2 34 °C (93 °F)
CPU #1 / Core #3 34 °C (93 °F)
CPU #1 / Core #4 38 °C (100 °F)
North Bridge 38 °C (100 °F)
South Bridge 40 °C (104 °F)

Voltage Values
CPU Core 1.42 V
+3.3 V 3.18 V
+5 V 4.94 V
+12 V 12.04 V
+5 V Standby 4.82 V
FSB VTT 1.47 V
North Bridge Core 1.49 V
South Bridge Core 1.14 V
South Bridge PLL 1.63 V
DIMM 2.30 V
DIMM VTT 1.15 V

[Dee]

Great. Many thanks sir.

[mikeyakame]

vPLL controls the phase locking between BCLK0 and BCLK1 for the CPU. Higher vCPU_PLL gives tighter control at high frequency but can also introduce extra jitter, and some skewing of clock signals between BCLK0 and BCLK1 (usually higher = advance, lower = delay). This is a brute method of aligning phases between different cores but it does work if you are patient enough and learn to look for minute differences visually. Misaligned phases in Vista for example can show up as delays to display refreshing, display drawn out of order, and so on.

[adamsleath]

all interesting.