Nehalem Overclocking @ XS

[dinos22]

woahhh I think I can get one for 10 euros with lesser bling

damn you guys
everything is cheap there
bloody LN2 is almost free, not to mention all the other stuff....if it wasnt so humid maybe it would be a nice place to move to hahahah

[EmineM]

I got mine from http://www.conrad.com/ for 30 Euros.
they ship to several European countries, im not sure if they ship out of the EU.

[oktay76]

good early bloomfield oc-results fugger, good work. finite u will get such results with an amd deneb too with ln2 of course....

[LardArse]

Why doesn't anyone admit that there is a high voltage cold bug with these cpus? Anyone who's benched them cold should know.

[HDCHOPPER]

ok sham eye believe ya .... but it's going to take a while for me to try lol

[Movieman]

Why doesn't anyone admit that there is a high voltage cold bug with these cpus? Anyone who's benched them cold should know.

Hey, it was cold in my office when I benched..
I took the room down to 68F from 72F..

[BenchZowner]

Hey, it was cold in my office when I benched..
I took the room down to 68F from 72F..

ROFL
That's still hot m8

p.s. Glad to see you're back Dave

[Fr3ak]

Either I am blind or the information about this runaway amp thing is not in this thread. How does it show practically?

I just finished my X58 roundup. I tested for max BCLK and measured CPU and memory voltage idle and load. Didnt notice anything suspicious, but when I used the 965 for WCG, it didnt produce a single wrong calculation for 16 hours then suddenly restarted, which is pretty odd as temps were even lower than the day before.

Surprisingly the MSI and Biostar board git the highest BCLK. Biostar up to 222 MHz for 1M, 218 MHz Prime 95. MSI hit 220 MHz Prime stable and 1M stable at the same speed. Asus boards were close. 215 MHz stable, 218 MHz 1M, Rampage II Extreme being slightly worse than P6T. Intel board sucked big time: 185 MHz 1M and 180 MHz Prime. The Gigabyte board refused to boot after making any changes in the BIOS with the 965, while it ran fine on 920. Bad sample I guess, didnt do any further testing.

I cant remember the performance numbers, but P6T was fastest on 32M. 1M was close on all boards. The Biostar was pretty fast for 3D, so was the RIIE.

[woffen]

What about the Evga X58, was that part of your tests?

[Fr3ak]

No, the EVGA is not ready yet, afaik.

[Sampsa]

Either I am blind or the information about this runaway amp thing is not in this thread. How does it show practically?

No you are not blind, I'm also hoping for some more information regarding this subject. In my tests I have measured 17,8A from ATX +12V yellow wires when CPU-test ran ok @ 4,644 GHz. Sometimes system crashed @ 16,3A. How can we benefit in overclocking observing the amps?

Seems like 3DMark Vantage's GT1 and GT2 are drawing very little power from PSU (4,5 GHz run):



12,14V * 3,8A = 46,132W

Then CPU-test 1:


12,07V * 16,8A = 202,776A

There is no time to react with ESC key and stop the CPU-test when system freezes and amps drop from ~17A to 6,8A or something.

FUGGER? Dr Who?

[The-Fox]

Update:

The uclk ratio I prefer to use is (the lowest possible) 1:2 of my ram speed (uclk runs at twice the speed of memory) and at lower speeds there are plenty more ratios to scale higher without running the sick memory speeds that I do, this is a huge gain in performance but for me it requires a lot of vmem and high end memory to run upwards of 2300Mhz (4700Mhz uclk) the memory controller has exceptional is its stability at this speed, I have not reached top speed and it will take even faster ram and more vmem to achieve that.

Fugger, is there any real world/synthetic improvements running the memory/uncore in higher speeds ?
Moreover, how does lower/higher timings affect performance, e.g., 7-7-7 vs 9-9-9 ? I wonder if the performance scale with better timing.

[Micutzu]

"Runaway" is generally en engineering term that describes a abnormal functioning of a part, running in a cause-generates-effect-generates-cause loop. For example, when current passes trough an electronic part some power is dissipated in the form of heat, equal to the part's electrical resistance times the square of the passing current. But the heat is increasing the part's resistance, so the power dissipated on it will increase, wich will make the part even hotter and making it dissipate even more power and so on, usually until either the part self-destructs or a protection system shuts it down. This is called "Thermal Runaway".

I am assuming Fugger noticed a point over wich the current drawn by the CPU increases rapidly because some transistors / junctions enter some kind of loop. There is one chance that it could be simple leakage (unwanted current flow trough theoretically-closed transistors), but i wouldn't bet on it since Intel's 45nm high-k process showed no such problems and the subzero temps should cancel them anyway pretty easy. The second guess, and my money is on this one, is that Nehalem's design is somewhat unfit for high voltages and over a certain value it just starts letting current pass trough it like there's no tomorrow. I have also noticed myself that over about 1.5V things stop scaling, from +50C all the way down to -80C, and this can only be happening due to electronic design.

[Jor3lBR]

"Runaway" is generally en engineering term that describes a abnormal functioning of a part, running in a cause-generates-effect-generates-cause loop. For example, when current passes trough an electronic part some power is dissipated in the form of heat, equal to the part's electrical resistance times the square of the passing current. But the heat is increasing the part's resistance, so the power dissipated on it will increase, wich will make the part even hotter and making it dissipate even more power and so on, usually until either the part self-destructs or a protection system shuts it down. This is called "Thermal Runaway".

I am assuming Fugger noticed a point over wich the current drawn by the CPU increases rapidly because some transistors / junctions enter some kind of loop. There is one chance that it could be simple leakage (unwanted current flow trough theoretically-closed transistors), but i wouldn't bet on it since Intel's 45nm high-k process showed no such problems and the subzero temps should cancel them anyway pretty easy. The second guess, and my money is on this one, is that Nehalem's design is somewhat unfit for high voltages and over a certain value it just starts letting current pass trough it like there's no tomorrow. I have also noticed myself that over about 1.5V things stop scaling, from +50C all the way down to -80C, and this can only be happening due to electronic design.

If confirmed and not 'solved' that's a big set back for OC.

[eva2000]

"Runaway" is generally en engineering term that describes a abnormal functioning of a part, running in a cause-generates-effect-generates-cause loop. For example, when current passes trough an electronic part some power is dissipated in the form of heat, equal to the part's electrical resistance times the square of the passing current. But the heat is increasing the part's resistance, so the power dissipated on it will increase, wich will make the part even hotter and making it dissipate even more power and so on, usually until either the part self-destructs or a protection system shuts it down. This is called "Thermal Runaway".

I am assuming Fugger noticed a point over wich the current drawn by the CPU increases rapidly because some transistors / junctions enter some kind of loop. There is one chance that it could be simple leakage (unwanted current flow trough theoretically-closed transistors), but i wouldn't bet on it since Intel's 45nm high-k process showed no such problems and the subzero temps should cancel them anyway pretty easy. The second guess, and my money is on this one, is that Nehalem's design is somewhat unfit for high voltages and over a certain value it just starts letting current pass trough it like there's no tomorrow. I have also noticed myself that over about 1.5V things stop scaling, from +50C all the way down to -80C, and this can only be happening due to electronic design.

If that's the case maybe the issue is specific to FUGGER's cpu and a select few, considering his cpus are on average clocking higher than most folks ??

[Gautam]

Why doesn't anyone admit that there is a high voltage cold bug with these cpus? Anyone who's benched them cold should know.

[Fr3ak]

Sounds more like a high voltage bug to me.

Does the (totally stupip, my opinion) Turbo-mode have something to do with the high voltage issues?

[Viss]

High volt when cold it seems.

[Micutzu]

If that's the case maybe the issue is specific to FUGGER's cpu and a select few, considering his cpus are on average clocking higher than most folks ??

Until more guys measure the current somehow we can't know but i am assuming all the chips have the same problem. Mine for example barely did 4.6Ghz @ 1.47V / -50C and the rig was pulling almost 400W in full 8-threaded load from the wall with only 3GB RAM, one HDD and one 8400GS VGA. Pretty high in my opinion considering the low core voltage, too bad at the time i tested i was unaware of the problem and didn't keep my eyes on the wattmeter all the time.

[Sampsa]

Hmm, maybe this whole thing is only affecting FUGGER's chip because the text he posted earlier looks like someone is describing the CPU for him (Fracois from Intel)?

"vcore ~1.5v, this is a "leaky" high IDV chip, scaling up with higher voltage is possible at a cost of runaway amp phenomena."

"your chip has a limit with all cores loaded that it will runaway on amps till it crashes." FUGGER's post

Francois explains here that there are Core i7 CPUs like the one he and FUGGER are going to use to do similar demo than AMD did with Phenom II:

"freak of statistic, this part is not in the middle distribution of the curve, it actually leak some serious current ... but it does OC, really, like hell." DrWho?'s post

If this is the case, we won't see runaway amp phenomena on "normal" ES and retail CPUs because those are not high leaking chips.

+ we also can't compare our results to FUGGER's chip since they aren't "leaky" high IDV chips picked by Intel

[Micutzu]

What i understand from Fugger's post is that there's a Vcore limit after wich the current just goes insane; the fact of the matter is that all the Core i7 results i've seen so far stop scaling around 1.5V, just like Fugger's chip.

Usual current leakage is a different matter though, I also noticed that high-leak parts (hotter chips) usually do well under subzero but it doesn't have anything to do with the 1.5V limit. Low temperature cancels transistor leaks and under LN2 the 1.5V limit, if it would be caused by regular leakage, would be moved much higher or even disappear completely. But instead it remains at the same value, wich proves it is NOT caused by electrons / voids jumping trough the transistor's insulating layer.

[Sampsa]

We will get video of this topic next week so stay tuned

[brolloks]

I'm new around here and thought I would post some of my experiences here. This is all on water cooling.

My thread over at OCF

http://www.ocforums.com/showthread.php?t=583975

[JPQY]

Great results you have brolloks

JP.

[Drwho?]

Yes, I am flying to Fugger's Light city this week, and I am taking some serious Kung Fu with me, from the crazy parts, to the super Kung Fu unlocking reserved for Manufacturers ...
The point I was to make is not the max frequency, it is more about fact that a Manufacturer demo is kind of pointless, out of the 100 000 Core i7 we already shipped the day of launch, it is obvious that some of them are just awesome crazy chips.
A manufacturer demo of extreme Overclocking is rarely representative to what the CPUs are on the market.
Since Conroe, I am very carefull not to use Picked processors (When I start controling the perf demos) ... I am going to show you what happen when you actually use picked parts.
The process technology people are actually able to understand what limit a CPU, so, if you look for it, you find a part that will be out of the distribution, but will still quality to sell. Those usually end up into the extreme edition and finish running at high frequency aircool into an OEM build machine, or in a Boxed Extreme Edition. Now, we of course sort them, and we got some extreme extreme parts.
Manufacturer of processors should stick to demonstrate what they are selling, not something that they cannot quaranty to its customer.
when I showed Skulltrail at 4.0Ghz, I knew that thermal cooling was the limits, and the QX9775 were all able to run at 4.0Ghz, if properly installed. (we did not want to quaranty it, for obvious reasons, but it was possible, the only impact was to decrease the life of the CPU)
If a processor does not always archive a frequency, I refuse to show it publically, it avoid to set up expectation that you can not follow up with a real product. Those are my guide lines, and I think everybody is fairly happy of it. I ll break the rule this week , but I want all of you guys to understand that it is not what a CPU maker should be showing. It is about setting expectation.

This baby is running at 4.8Ghz aircooled, without pushing voltage much:

I know that this baby will fly, because mathematically speaking, it shows the right caracteristic.

I am waiting 5 mores like this Monday, directly from the men who sort them.

(Notice that I go super honnest on the disclosure side)