Why some CPU's OC better than others?

[BoboKatt]

Hi there and please excuse my ignorance but I am having some issues wrapping my head around why two exact CPU's (in this case C2D e6600) could OC so differently.

I originally bought a 29A e6600 a while back. I am using an eVGA 680i mobo (the first revision). I can get around 3.2 GHz with no issues and a moderate increase in Voltages. However to get to the magic 3.6 GHz it requires an insane amount of V. From reading many many posts on this site and ones like it, you can easily see that OC'ing is a hit and miss and nothing is guaranteed.

My question is more on the lines of simply how is it that 2 e6600's could be so different from each other. My friend came over on the weekend and brought his e6600. We went through the hassle of taking out my Tunic Tower and taking out my CPU and placing his in there. We applied the AS5, replaced the HT/Fan and away we went.

His CPU hits 3.6 GHz on my mobo/RAM combo with my same voltages that i need to keep my e6600 at 3.2 Ghz. I simply don't get it. Are the chips different physically?? I would love someone with way more tech info than I have to help me understand this. We talk about week and batches but his is the same A29 as mine. We both bought it from the same online retailer at roughly the same time.


I could understand if his was a different chip, say an e6700 or a e6400 or maybe he had a better board or different PS etc etc. But to eliminate these possible differences we made sure to use my system and just change the CPU. What exactly does Intel and AMD do to these chip, chip by chip I mean? Better material?? Or just simply luck of the draw?

Don?t get me wrong, at the end of the day when a CPU you buy that runs at 2400 Mhz OC?s to 3200 Mhz and we ?complain?, it really is rather funny but it just intrigues me.

 

[f4phantom2500]

well think about it, there are millions and millions of transistors on that little tiny chip. when they make chips, they don't have group 1 making e6600's and group 2 making e6700's, etc etc. they make all of the chips the same way. then they test them to see how they run at whatever speed. the ones that do best get sold off as the highest end chips, and the worst get sold as the lowest end ones. however, sometimes, they make a batch that's really really good, so most of the chips have a lot of headroom and can overclock well (after all they can't afford to sell every chip as the highest end one, they wouldn't sell very many that way). of course, that also means that some batches are really really bad, and most of them don't overclock well.

so you see, even though they have the same model number, 2 e6600's are just as similar (in terms of their overclocking limit) as an e6600 and an e6700. that is, they aren't. each chip is different.

get it?

edit: just because they're from the same batch too doesn't necessarily guarantee a similar overclock from both either, however it does make it more likely, which is why people try to get chips from specific batches. i mean if the batch is good overall, you want one because youd have a better chance of getting a good one. however not all of them are good, people just weigh the odds. the opposite is true too; a chip from a "bad" batch isn't necessarily a bad overclocker, there's just a better chance of it.

 

[trOver]

bingo

 

[MarcVenice]

I also heard that the wafers the chips come out from also play a role. It is believed that chips coming from the middle of the wafer, will have the best characteristics, and will thus often OC better. Now, if lets say 1 batch comes from one and the same wafer, the ones in the middle will probably OC better.

Also, the reviews we are seeing right now about the e6750 are probably done with chips that have been handpicked by Intel, and for us lowly customers it might be very hard to get a e6750 to 4ghz, at very low temps and with minor voltage increases like they can on the reviewing websites. I don't know how true this is though, but I've read it many times before, and have seen it happen with new cpu's coming out.

 

[f4phantom2500]

Originally posted by: MarcVenice
I also heard that the wafers the chips come out from also play a role. It is believed that chips coming from the middle of the wafer, will have the best characteristics, and will thus often OC better. Now, if lets say 1 batch comes from one and the same wafer, the ones in the middle will probably OC better.

Also, the reviews we are seeing right now about the e6750 are probably done with chips that have been handpicked by Intel, and for us lowly customers it might be very hard to get a e6750 to 4ghz, at very low temps and with minor voltage increases like they can on the reviewing websites. I don't know how true this is though, but I've read it many times before, and have seen it happen with new cpu's coming out.

it doesn't even matter; the end result is my explanation

 

[StopSign]

The E6750 that Anand just tested with is probably the best E6750 that will ever exist.

 

[theprodigalrebel]

Yeah, remember the Core 2 Duo launch article, where they got a E6600 to 4Ghz on a Tuniq Tower?

Enthusiast Forums > Hardware Sites when trying to form a ball-park figure of what to expect from a CPU (in terms of OC potential). Just imagine if all of us could run E6600s at 4GHz on air-cooling? Sweet mother of all that is holy! :Q

 

[BoboKatt]

Yup all that makes sense. I was thinking maybe quality control had something to do with it but obviously that cannot be it. I can now "see" that when you are dealing with millions of transistors etc, there will be chip to chip variations no matter what.

Yah I've read many reviews and the recent one on the e6750 made me drool for its inherent OC ability but as some have stated here LOL that might as well be the best chip anyone will see.

Would this explain why some are selling Engineering Samples of come CPU?s online on eBay for fortunes? These were maybe hand picked?

Also just a last questions? is there any physical difference between an e6600 and e6700 in transistor count etc? In other words are these the same chips just that they were tested and the e6700 simply runs better? And does this apply to the X6800 except those are unlocked I guess but again basically the same chip that came out of the process with better results?

 

[F1N3ST]

Nah, the new stepping G0 runs very cool, with not many volts, so they probably are cherrypicked but will be VERY common to get lower temps at higher speeds.

 

[MarcVenice]

I think bobo, that your last question has you assuming correctly.

The engineering samples WILL oc better though, and I think are handpicked, and that's why they go for a fortune. I wish I could get my hands on of those...

 

[cprince]

Not all chips are created equal. The manufacturing of chips is a very complex and complicated process with many variables. Manufacturers often give lead ways(tolerances) in order to get higher yields. This lead way is what allow you to overclock. Depending on where the chip is on the wafer and where the wafer is on the rack determines the amount of tolerances.

 

[f4phantom2500]

bobo: yes, every c2d is the same chip actually. even the 2mb models. see, sometimes the 4mb cache has problems, but at least 2mb of it isn't defective. so they disable the extra cache and sell it as a lower model. even if they aren't defective sometimes they have to disable some of the cache and sell the chip, just so they'll be able to meet the demands for each processor. there is no way to unlock this cache. and yes, even ones with unlocked multi's are the same. generally speaking, ones with higher tolerances (explained by cprince) are sold as higher end chips, but sometimes the design of a chip and the manufacturing process is so good that even the low end chips of the series have really high tolerances, which is basically the case with these core cpu's; even the chips that "suck" at overclocking can get at or pretty close to 3ghz, and at that speed will beat anything from the amd camp, even when they're overclocked too (a64's cap out at about 3ghz).

 

[BoboKatt]

Wow great info and very easy to understand. Exactly what I needed :0 Thank you all.

 

[theYipster]

Indeed. It's just like baking cookies

 

[Zap]

Originally posted by: BoboKatt
My question is more on the lines of simply how is it that 2 e6600's could be so different from each other.

The better one had some sacrificial goat blood used for heatsink compound the first time it was booted up.

 

[aigomorla]

board and ram also play a big role too.

Im starting to love my F batch xeon 3220 after i dropped the gigabyte P35 on her.

680i is absolute CRAP when you compare it with the P35. And the P965's are looking a lot smaller in comparison also.


Incase soemone is wondering how good this P35 is, i can scale my xeon all the way up to 460fsb x 8.

on the 680i, i was hitting a wall at 375x9 and 415x8

 

[BoboKatt]

aigomorla -- This is very interesting what you are saying about the 680i. One thing we did was to take the same e6600 and on my eVGA 680i, we lowered the X to see what the max FSB would have been on both. I presumed that essentially by doing this I would be able to isolate the highest FSB the board would handle independent of the CPU.

What was interesting was that I was able to get a higher max FSB on the board when I was using my friend's e6600. This makes so little sense to me as I again assumed that it would be the board's max, once you eliminate all other components, lowering the multiplier and setting RAM speed and timings really really low.

Humm so.... P35 are looking good eh... which one? Christ here we go again... more mobos and chipsets to watch and follow for months before making an "educated" guess LOL.

 

[Diogenes2]

When you think about it, it's kind of funny, wondering why CPU's sold to run at 2.4g will or will not run at 3.6g .....

 

[DasFox]

How far do you guys think we are away from technology that will produce consistent quality without any variation? 100% identical in performance and life. Is this possible in the next 10 years?

ALOHA

 

[Fenixgoon]

well consider the electrical conductivity in a crystalline material - heavily dependent a limited number of defects and dopants within the crystal structure. i don't remember the equation off hand (i can look it up if you really want... hell, i may just to remember). anyway, impurity levels for semiconductors range somewhere in 10^-5, i believe. so let's say you have some pure silicon, which is 28g/mol. so in the 6.022x10^23 atoms you have for 28g of silicon, you have something like 6x10^18 impurities. since these impurities can be dopants, we might actually have a greater conductivity than pure silicon, despite a deformed lattice.

and while i may sound like i know a lot... i really don't, that's just basic electrical properties of materials stuff. time to look up that conductivity equation now

 

[DasFox]

I guess time to start looking for that material of the future, which doesn't sound like silicon anymore.

 

[wwswimming]

Originally posted by: DasFox
I guess time to start looking for that material of the future, which doesn't sound like silicon anymore.

Rubylith ?

just kidding, that was what we called the red tape we used to
cut microwave circuit artwork, circa 1985.

i don't understand the very minute differences that can
cause the circuit connections in a CPU to OC differently.

at the board level, little things, like a tiny goober of solder,
at microwave frequencies it becomes a circuit component,
with capacitive & inductive properties that can change
transmission line characteristics.

having worked around microwave circuitry at the component
& board level, for example watching another engineer using
pieces of styrofoam coffee cup to tune a coupler, well it
made it obvious that at microwave frequencies, almost
every little detail of circuit geometry ~and~ chemistry
will affect circuit performance.

not just dielectric constant (epsilon), but the epsilon "prime"
and "double prime" (material properties change with frequency
as well as temperature).

the same effects are applying on a Core2Duo. not having
worked in chip fab beyond slice/dice & die attach, i would
only be guessing.

pretty impressive these chips are though.

 

[CTho9305]

Originally posted by: DasFox
How far do you guys think we are away from technology that will produce consistent quality without any variation? 100% identical in performance and life. Is this possible in the next 10 years?

ALOHA

It's currently getting worse rather than better, actually. As devices shrink, the variation goes up a lot. A very simplified explanation is this: transistor speed is strongly affected by threshold voltage (the voltage at which they turn on/off). The threshold voltage is largely determined by the doping of the silicon (what kind of non-silicon atoms and how many there are). With very tiny transistors, the number of dopant atoms can be in the low thousands or even in the hundreds, and they're deposited randomly. Imagine a grid, and try throwing sand onto it so that each cell has about the same amount. If you want to put 50lbs of sand in each cell and the cells are big, it's pretty easy to get 1% accuracy. If you want to put 50 grains of sand in each cell, you're going to have a wide distribution in the number of grains per cell. If the grains were dopant atoms in a transistor, you'd get a wide distribution of device speeds.