How high of vCore is too high for Core2Duo? 1.35v, 1.40v, 1.45v?

[BlueAcolyte]

So what do you guys think is a safe VCORE stat for an E4500 at 1333FSB? (333MHZ) I was only going to push to 1066FSB. (266MHz)

Would stock VCore be enough for 266MHz, and what would i need for 333MHz?

 

[Modelworks]

I've got an E4500 running at 300Mhz on 1.34 volts.
Anything below that and it was stable at first but after about an hour would fail orthos test.
I can get it to 333 at 1.41 volts, but its not worth the extra heat and the extra increase in speed really isn't noticeable for me.

At 1.34 Vcore its running at 31C idle and reaches 52C under full load, Arctic Freezer cooler.

 

[BonzaiDuck]

The E4500 is binned for running at 800 Mhz FSB.

My quad Q6600 is rated for 1066.

Now, the two processors are not comparable, because there are more cores packed onto the B3 Q6600, it's farther up the list of performance in the C2(x) model line, anyway, probably won't over-clock PERCENTAGE-WISE as much as this Allendale. But I have a better start, because the processor is guaranteed to run stock at 1066, and my motherboard is rated to run stock with processors rated at 1333.

I've got the experience of running an 800 FSB processor on a board rated for 800 FSB, and I over-clocked the system to 1000 FSB. This was too much for the motherboard. Since then, the mobo manufacturers have heard some sort of market signal to build boards capable of running at faster FSB speeds.

I've said too many times that there's a reason intel picks a voltage as its retail-box "maximum." I said this is also an indication that the voltage could be pushed higher -- safely. But how high is a matter of individual judgment. In the case of my old Prescott, it wasn't the CPU that was threatened by pushing to only a 25% over-clock -- it was the motherbaord.

But here, you're looking to push an 800 Mhz-FSB processor to 1,333. That's a 66% over-clock.

It's not for me to say "what's a safe OC level," or what's a "safe VCore."

Anyway, given my last post, I spoke my mind here, and that's enough for me.

 

[BonzaiDuck]

I'll offer this, since my "Maximum PC Magazine" just arrived in the mail. [As an aside, let me say I subscribe to this magazine with a tongue-in-cheek attitude. Their reviews always seem to favor companies and products with full-page color ads appearing at the end of a comparison review, and they always make sure to exclude products that show superior performance to the "Kick Ass 10" -rated winner of the comparison.]

This month (well, January 2008) -- features "OverClock Your PC" beginning on page 20.

On page 28, they list a table of over-clocking results for four INtel processors, with speed, voltage, FSB x Multiplier, and CPU-revision:

C2Q Q6700/Kentsfield: stock: 2.66Ghz OC'd to: 3.44 VCore: 1.40 FSBxMult= 343x10
revision: B/G0

C2D E6300 Conroe: stock: 1.86 OC'd to: 2.88 VCore: 1.50 FSBxMult= 412x7
revision: F/B1

C2D E4500 Allandale: stock: 2.2 OC'd to: 3.3 VCore: 1.40 FSBxMult= 300x11
revision: D/M0

Pentium E2160 Allendale: stock: 1.8 Oc'd to: 3.37 Vcore: 1.56 FSBxMult= 375x9
revision: 2/L2

Bench configuration: EVGA 680i SLI, 2GB Corsair Dominator, PCP&C 1KW PSU, air-cooling


PLEASE NOTE: In the TEXT of the article, their " . . . for example, . . " settings are much more conservative, or less than a 10% Over-Clock.

Earlier literature I've seen, as with "CPU Power-User Magazine," notes that a 10% OC is essentially risk-free. Given the elastic behaviors of the new Intel 65nm processors we've seen, we wouldn't know -- really -- how the risk now changes -- given that a 20% OC on these processors is so easy that a NOOB can do it -- "So easy a Cave-Man can do it . . save big money on your processor insurance . . . "

 

[GundamF91]

Originally posted by: BlueAcolyte
So what do you guys think is a safe VCORE stat for an E4500 at 1333FSB? (333MHZ) I was only going to push to 1066FSB. (266MHz)

Would stock VCore be enough for 266MHz, and what would i need for 333MHz?

See my other post on the "factory default vCore" issue and on how to find out about what your default value is. I am sure now that the key to overclocking potential is linked with the default vCore or VID that Intel puts on the processor. Each is processor is "soft locked" so to speak at a particular VID at the default max multiplier. The motherboard is able to change that, if your BIOS allows it, such as Abit IP35E can increase vCore as well as reduce mutliplier.

My current E4500 vCore defaults at 1.30v, I've also tested one that defaults at 1.325v. I've tested them multiple times, and everytime the 1.30v won out in terms of stability and temperature. So if your processor is better than average factory default vCore for its class, then you've got a winner.

As for your question, the FSB it'll hit is a function of processor AND motherboard. with most P35 board, the standard is 1333mhzFSB or 333mhz. As for the processor, it's not a matter of what FSB it'll run at, it's more of the overall processor speed. you can do 200x11=2.2Ghz, or you can do 300x7=2.1Ghz. Similar performance CPU wise, but your overall computing performance is better at higher FSB. You can run 333mhz if you drop your multiplier down enough to allow for higher FSB.

I've tested the 1.30v E4500 thoroughly enough to conclude that I can use the factory default 1.30v for vCore while running at 333mhz. this translates to 1333 quadpumped FSB. The speed of the processor does hit a ceiling at about each level of vCore. With 1.30v, my E4500 can do 333x9=3.0Ghz. It idles around 22c, and load temp hits 48c, with water cooling. This is more than satisfactory for me with 36% overclock, and also does not impact the processor much since I kept at the same vCore.

I have not established a firm vCore for E4500 for hitting 333x10=3.3Ghz. It appears that 1.40 will do it for 30min Orthos, but my vDroop lowers it enough that I think it'll take 1.42v to officially be stable. And as I mentioned above, I can also drive it up to 1.6v to make nearly 3.7Ghz on the E4500, but that's really pushing it so I've stopped doing it now.

 

[BlueAcolyte]

Ah, so that means that running at 333MHz with a full multiplier is a no-no. I am going for 266 MHz x 11 = 2.93 GHz, I suppose it depends on what my stock vcore is (mobo, gfx card, and psu havent gotten here yet, guess i have to wait thanks for advice)

 

[moparacer]

Well, with all of the overclocking peeps in this forum and at XtremeSys...com., how many of us have fried an intel core2 duo / quad to date? How many people have reported an overvolt-failure? Curious?

Me too....I want to see how many have fried a cpu with 1.6+ volts.....I have tesed mine at MAX OC and primed for several hours at 1.6 and it didnt explode in a ball of flames...

I have my 2180@ 1.52 now and dont really expect it to die a violent death anytime soon...

And that Vcore on the 2160 that MaxPC tested is right in line with what I have found on my 2180 to do 9X375

 

[BonzaiDuck]

Originally posted by: moparacer

Me too....I want to see how many have fried a cpu with 1.6+ volts.....I have tesed mine at MAX OC and primed for several hours at 1.6 and it didnt explode in a ball of flames...

I have my 2180@ 1.52 now and dont really expect it to die a violent death anytime soon...

And that Vcore on the 2160 that MaxPC tested is right in line with what I have found on my 2180 to do 9X375

Well, you probably won't keep this setup longer than a couple years, if you're as much of a hotdog enthusiast/over-clocker as the rest of us.

My personal preferences are these:

* If I were to run my Q6600 at 1.5V, I'd only want to do it briefly, taking advantage of the "OC Profiles" in my BIOS when I play games.
* If I wanted to find an OC setting at which I wouldn't occasionally worry about some sort of eventual failure -- a fixed OC setting -- I'd look at settings between the "spec maximum" or the stock "default" and 5% over that spec.

Overall, my concern is not just about a low-end or low-price processor failing, but about my data and the inconvenience of replacing the processor under "surprise" conditions. For instance, if I had an E6750, I suppose I wouldn't care about needing to replace it in 6 months to a year unexpectedly due to failure.

That latter *-ed "rule-of-thumb" is just self-imposed, but others here have noted that "safety" depends on how far you departed from the "default." My 25% over-clock setting is actually a few percent less than the default in Vcore -- and at the CPU's stock multiplier.

Even under these cautious choices, I probably would replace this processor in less than two years. In fact, I'll either replace it for G0 stepping, a dual-core G0 E6850, or a Penryn dual-core in a matter of months.

 

[aigomorla]

Originally posted by: BonzaiDuck

Originally posted by: moparacer

Me too....I want to see how many have fried a cpu with 1.6+ volts.....I have tesed mine at MAX OC and primed for several hours at 1.6 and it didnt explode in a ball of flames...

I have my 2180@ 1.52 now and dont really expect it to die a violent death anytime soon...

And that Vcore on the 2160 that MaxPC tested is right in line with what I have found on my 2180 to do 9X375

Well, you probably won't keep this setup longer than a couple years, if you're as much of a hotdog enthusiast/over-clocker as the rest of us.

My personal preferences are these:

* If I were to run my Q6600 at 1.5V, I'd only want to do it briefly, taking advantage of the "OC Profiles" in my BIOS when I play games.
* If I wanted to find an OC setting at which I wouldn't occasionally worry about some sort of eventual failure -- a fixed OC setting -- I'd look at settings between the "spec maximum" or the stock "default" and 5% over that spec.

Overall, my concern is not just about a low-end or low-price processor failing, but about my data and the inconvenience of replacing the processor under "surprise" conditions. For instance, if I had an E6750, I suppose I wouldn't care about needing to replace it in 6 months to a year unexpectedly due to failure.

That latter *-ed "rule-of-thumb" is just self-imposed, but others here have noted that "safety" depends on how far you departed from the "default." My 25% over-clock setting is actually a few percent less than the default in Vcore -- and at the CPU's stock multiplier.

Even under these cautious choices, I probably would replace this processor in less than two years. In fact, I'll either replace it for G0 stepping, a dual-core G0 E6850, or a Penryn dual-core in a matter of months.

Totally agree with you duckie.

Heres my basic rule in determining what the max safe voltage/heat is on a processor.

Basically your fighting 2 things when you overclock:

1. voltage
2. heat

Now the way chips work, the colder you can get your processor, the easier the overclock is. That is why people on phase usually get 200-300mhz higher then people on regular cooling. Sub ambients allow greater clocking at lower power draw.

So for air i would look at it this way:

1.4-1.42 is absolute max, unless you have a cold room where the ambients dont get very hot. Then i would look at cpu temp. Determine how long you intend to keep the chip. A Intel engineer once said every 10C you lower your chip in load temps, you effectively doubled the life of your processor.

The MAX safe temp for any C2D or C2Q should be 70C. If your riding on 70C id call that safe, if your near 75C your asking for too much.


If you guys are on water, people can do 1.5V and have there chips last longer if not as long as people on air running at 1.42 Remember the 10C rule. But at 1.5V you risk electron migration damage. I'll go for the sucide runs at 1.6V but anything higher and i'll be asking for it on my water kit.

1.6V+ your looking at phase territory with sub ambient load temps. Were talking about -10C LOADED.

 

[CTho9305]

Originally posted by: MetaDFF

Originally posted by: tenax
what i'm curious about is is it the voltage..or the operating temperature that kills? i believe it's the voltage unless the temperature is extremely high. (like years ago when i forgot to put the heatsink on my athlon 800 and in 15 seconds it was literally toasting.

Well both a high voltage or a high operating temperature can kill a chip in different ways.

If the operating temperature is too high, the current density through the metal interconnects can cause the metal to be "burned off" in a process known as electromigration. Most chips today are designed to handle the current density up to 125 C before electromigration starts to occur.

Got a source for that 125C claim? Is that 125C at the transistors, at the outside of the silicon die, at the outside of the chip package, or as reported by the monitoring software? Whether or not it's true, running at high temperature is bad because it speeds up other wear-out mechanisms like NBTI. See also my post here.

 

[aigomorla]

Originally posted by: CTho9305

Originally posted by: MetaDFF

Originally posted by: tenax
what i'm curious about is is it the voltage..or the operating temperature that kills? i believe it's the voltage unless the temperature is extremely high. (like years ago when i forgot to put the heatsink on my athlon 800 and in 15 seconds it was literally toasting.

Well both a high voltage or a high operating temperature can kill a chip in different ways.

If the operating temperature is too high, the current density through the metal interconnects can cause the metal to be "burned off" in a process known as electromigration. Most chips today are designed to handle the current density up to 125 C before electromigration starts to occur.

Got a source for that 125C claim? Is that 125C at the transistors, at the outside of the silicon die, at the outside of the chip package, or as reported by the monitoring software? Whether or not it's true, running at high temperature is bad because it speeds up other wear-out mechanisms like NBTI. See also my post here.

Holy cow, your post on EM is Godly... you should make a general one, and have them turn it into a sticky.

Also EM is a reason most people on XS believe that kills today's DDR1066 ram sticks. Vendors are okeying 2.2V+ on a 1.9V chip which leads then to fail. I personally lost 3 pairs of my tracers.

Another place where people think EM is the killer is boards. To make a quadcore run ata 1600FSB+ you need to push a lot of votlage in the north bridge. Over long periods of time, the north bridge will leak, and make your overclock unstable, and then eventually die a horrible death.

Ive lost 3 EVGA 680i's because of this. And they were all watercooled. :


I think i basically got down which vendor has greater long term stability. And im a bit wiser on the voltage. I already mastered the heat issue.

 

[CTho9305]

you should make a general one, and have them turn it into a sticky.

On what? I've thought about writing about the failure mechanisms most common in OCing, but most of my knowledge is theoretical. I don't know for sure which failure mechanisms really kill chips most often. Unfortunately I can't share the data I look at (e.g. analysis of damage to real silicon under controlled voltage/temperature conditions across wide ranges), and I haven't had much luck finding publicly-available data. There are forum members here who know a lot more about chip failure than I do, but they're likely in the same situation when it comes to providing actual data.

 

[BonzaiDuck]

AigoMorla !!

I need to consult!

In your water-cooled marvel, I gleaned last summer that you were running Tracer DDr2-1066 (or 1000's?) at the full spec -- probably a divider <> 1:1 -- with latencies 4, [x],[y], 8 or 9. I think you'd posted that you were running them at 2.2V

In order to get super-tight latencies 1:1 with lower FSB, I had to twist up the voltage from 2.15 to 2.2. BIOS monitor shows 2.24V.

I felt pretty sure that the 2.2V was within or at the warranty-spec on these. NefariousCaine told me he'd had a few pairs that he'd run at 2.3 and 2.4V, but I wouldn't venture into that territory myself. He also had one pair with a stick that failed.

Are you saying that setting them at 2.20V is excessive?

 

[aigomorla]

Originally posted by: BonzaiDuck
AigoMorla !!

I need to consult!

In your water-cooled marvel, I gleaned last summer that you were running Tracer DDr2-1066 (or 1000's?) at the full spec -- probably a divider <> 1:1 -- with latencies 4, [x],[y], 8 or 9. I think you'd posted that you were running them at 2.2V

In order to get super-tight latencies 1:1 with lower FSB, I had to twist up the voltage from 2.15 to 2.2. BIOS monitor shows 2.24V.

I felt pretty sure that the 2.2V was within or at the warranty-spec on these. NefariousCaine told me he'd had a few pairs that he'd run at 2.3 and 2.4V, but I wouldn't venture into that territory myself. He also had one pair with a stick that failed.

Are you saying that setting them at 2.20V is excessive?

i do a lot of folding. And it was 4-4-4-8 Dont have those settings anymore. Im done benching for records.

Loading up the ram 24/7 @ 2.1Vcore lead to both sticks dying a horrible death.

Right now on my 4th pair im running them at 1.95V If these die again, im going OCZ. :X



Basically too many ram sticks dying on micron D9's means the IC's arent that great. There awesome overclockers, but there like cheetahs. They run out and die. What we need are throughbreads. Something that can go on for the long haul.

CTHo:
The thead should be on dangers of pushing the cpu too much voltage wise. Theres a BIG BIG MISCONCEPTION thats going on in this forum. People think as long as your temps are in check, the voltage doesnt matter. People need to see reasons on WHY pushing too much voltage even with excellent temps is dangerous.

 

[BonzaiDuck]

I agree. That was a prevalent myth around here for some time, and even I had bought into it. But I had this intuitive loss of comfort because I noticed a sort of pattern -- even air-cooling -- when you bump up the FSB in linear, equal increments, and you see the voltage requirement going up 2, 4 or more increments as it moves forward. This also seemed to introduce a threshold where temperature was noticeably increasing.

I'm still running the Crucials at 2.2. I dropped back from a very stable 3.2Ghz settings because I would've needed to loosen the latencies to push the FSB up -- or -- I could only run a 2T command-rate at 3.2. But dropping back 12 Mhz in two steps, I could get to 1T with very good increase in bandwidth. So I finally settled on 3.15 Ghz -- leaving the Vcore at 1.4125. Now I've discovered I can drop the VCore back to between 1.38 and 1.39, and it's still stable, and cooler. I've got it within 2 Mhz of a setting where I could run the memory at 2.15 with slightly looser settings. If I can drop the voltage back to 2.175, it will be even better.

 

[BonzaiDuck]

Here's a hypothesis, though:

Different modules of DDR(1) that I've had behave differently. You could buy OCZ Gold EL DDR-500's, and at that speed, you could set the latencies at the stock value. If you tried to drop the speed to 480, 460, etc. and loosen the latencies, you couldn't do it.

Then there were the Platinum DDR-400's. You could keep the latencies at 2,3,2,5, increase the speed to 452 at one notch below the recommended maximum voltage.

So I'm thinking that DDR2-1000's had BEEN at the upper limit of that technology last year. I think the rated latency settings were 5,5,5,15 at that speed, while they would default to 5,5,5,18. But maybe they were "specially-binned" DDR2-800's. So if you tightened the latencies the way you did, and ran them full-bore at the rated speed in Mhz with the voltage set to the max, maybe it stressed them too far.

Now -- I've got these running as "DDR2-700." The latencies are not tightened down as far as NefariousCaine has shown they will go, but the settings are really tight -- at 1T command-rate. I ALSO have to run the modules at 2.2 -- hopefully a notch less -- but let's assume the worst scenario and I have to run them at 2.2.

So, like Clint Eastwood says to the bad guy in "Dirty Harry" -- "Do you feel lucky? . . . . . Well, do you, ____?" and I'll refrain from using the word he called the bad-guy, because I'm really talking to myself . . . .

 

[BonzaiDuck]

CORRECTION: per DDR-500's: " . . . If you tried to drop the speed to 480, 468, etc. and TIGHTEN the latencies, you couldn't do it."

 

[The-Noid]

1.45v QX9650.

 

[BonzaiDuck]

AigoMorla's experience and observations are pretty sensible.

Even the reseller spec on the QX9650 is low at just below 1.36V.

Some others here have surmised that if you go much beyond the monitored "default" voltage on these processors, troubles due to over-volting them will -- in the LONG RUN -- shorten processor lifespan. Now -- that could apply to the load voltage more than the set or monitored idle voltage. There does not appear to be any hard and fast certainty about it, although it makes some sense. 1.45 . . . . IS . . . . only 6% above that level.

But I notice that with one setting, Yoxxy has set his Northbridge to 1.63V. If he's using a water-cooled setup for the processor, what attention has he paid to cooling the NorthBridge, and what exactly ARE the less-risky limits for Northbridge core voltage?

 

[AVP]

aigomorla 1.42 is not the absolute max for air. 1.5 is.

 

[aigomorla]

Originally posted by: AVP
aigomorla 1.42 is not the absolute max for air. 1.5 is.

on a quadcore?

1.5VCore and your pushing a little below 300W of heat on that CPU.

What aircooler do you know that effectively disapates 300W of heat @ 22-25C Ambients?


Here incase you dont believe me:

http://www.xtremesystems.org/f...howthread.php?t=152222


And did you not read CTho's excellent post on Electron Migration? God seriously whats up with people thinking its okey to push high voltages on these chips.

Please DONT be telling people its okey to push voltages in the 1.5Vcore ranges.

This is exactly the reason why i say CTho should make his EM post as its own warning people of using too much voltage.

 

[MetaDFF]

Originally posted by: CTho9305

Originally posted by: MetaDFF

Originally posted by: tenax
what i'm curious about is is it the voltage..or the operating temperature that kills? i believe it's the voltage unless the temperature is extremely high. (like years ago when i forgot to put the heatsink on my athlon 800 and in 15 seconds it was literally toasting.

Well both a high voltage or a high operating temperature can kill a chip in different ways.

If the operating temperature is too high, the current density through the metal interconnects can cause the metal to be "burned off" in a process known as electromigration. Most chips today are designed to handle the current density up to 125 C before electromigration starts to occur.

Got a source for that 125C claim? Is that 125C at the transistors, at the outside of the silicon die, at the outside of the chip package, or as reported by the monitoring software? Whether or not it's true, running at high temperature is bad because it speeds up other wear-out mechanisms like NBTI. See also my post here.

To my understanding the 125C temperature is the metal interconnect temperature (i.e. ambient temperature is 80C and local temperature is 125C on the interconnect). The process design kits I've seen spec metal current densities up to 125C for continuous reliable operation for something like 10+ years. Sometimes the "failure" specified is incredibly stringent. IIRC the last design kit I looked at spec'ed the poly resistor current density at 125C for reliable DC operation for something like 15 years such that the resistance deviated by only 1% (I could be wrong, it's been a while).

Now say this resistor was in a consumer processor and you are pushing more than the spec'ed current density, maybe the resistance will deviate by 1% in something like 7.5 years, but it won't matter because most of us won't have that same processor by then. But for something like a sub-oceanic fiber repeater that will operate for 25+ years, that reliability spec starts to really matter.

I agree when you say that it would be hard to write an article on the failure mechanisms common to OC'ing since there are so many possible different failure modes. Most of these failures happen over long periods of time with a cumulative effect. Without actually looking at the failure model that Intel uses to predict their transistor reliability (hence the spec'ed "safe" Vcore), it would be impossible to say what is a safe Vcore over-voltage to ensure reliability of n number of years.

 

[BonzaiDuck]

MetaDFF -- Per your last paragraph, we are entirely in agreement -- entirely.

The only "data" we have are incidents of failure reported here -- but never tallied. I supposed we'd need honest participants who could cite the voltage applied constantly, the length of time at that voltage before that failure, and several other facts. AigoMorla mentions something else that is highly relevant: usage. He notes that he has run his processors in "folding@home" 24/7 over some unspecified time. Somebody else may play a game or two, let the system remain idle for periods of time, do some word-processing, etc. -- and never stress the processor the same way.

The point I made elsewhere is that there is an interface between sound business practice and engineering -- related to a "failure model" such as you cite, test results on large samples of product, and a desire to reduce unnecessary costs (like RMA returns.) So there is a DATUM -- what the manufacturer says is a "maximum voltage" for which they will honor a warranty agreement. It's not much to go on, but it isn't some hocus-pocus number pulled out of a hat, either. You'd figure that their choice of that number reflects an expected value of returns-under-warranty at near zero. Above that, the probability of failure, over some unspecified but fixed amount of time, increases.

Maybe it is a gradual increase, a linear increase, a non-linear increase. We can only guess. And only the manufacturer and his testing lab would know for sure.