Q6600 Core Layout?

[Syzygies]

Does anyone recall what document at Intel authoritatively specifies how cores are numbered in a Core 2 Quad?

I've made the following collage to help picture what's going on inside the processor:

Quad Core Core Layout

The left image is my motherboard, oriented as it sits vertically in my case. The yellow highlighting marks the side of the processor that spans between the two 775 notches. Note that this is the side that runs parallel to the hooked clip that clamps the 775 socket closed, which is the bottom edge as oriented in my case. The right image shows the recommended way to apply AS5 to an Intel quad core, along the red line.

What I don't know is whether Core 0 is nearest to the bottom, or nearest to the top. I've seen assertions in other forums, but they're undocumented, and I'm not inclined to take them on faith.

My "Core 0" runs hotter than my other cores, after lapping my Q6600 and TRUE. The overclocking potential is determined by the worst core. This always has been my hottest core, so this could be beyond my control. Or, by relapping and/or fiddling with TRUE screw pressures, I may be able to somewhat even out these temps.

If one gets past absolute flatness as a religious tenet (let's leave shine out of the discussion), there may be a germ of truth to Thermalright's recommendation not to lap. Their convex bow may be intentional, like the water block inserts to intentionally bow cpu water blocks. And studies of hand lapping show that one always ends up with a slight bow. If I must end up with some very slight bowing after lapping, I'd want it to tend to a cylinder that made contact along the AS5 application line. This argues for favoring certain hand positions and sanding directions as one laps the TRUE, so that the inevitable errors break favorably to encourage this axis of bowing.

The TRUE is heavy. In the past, I've seen Cores 0 and 3 run hot, or Cores 0 and 1 run hot, but rarely any other combination. This has to be a giant clue. If the weight and bowing of the TRUE interact to vary the pressure over time, this makes sense: Either the same half runs hot, or the outside edges run hot.

But, which half? I'd guess that the bottom half is favored by the weight of the TRUE, who knows how I adjusted my screws?

I'm certain that it would help me to understand this diagram and its implications, before relapping, rotating my TRUE, or tweaking screw pressures.

 

[Drsignguy]

This is Perfect! I have been wondering the same thing? Great job Syzygies! I have a Tuniq 120 lapped/ cpu lapped and I have 2 that run a bit warmer. Core 0, core 1. This will be some help. Again good job!

 

[imported_ST]

Just to play devil's advocate here, but after owning / playing with 4 Yorkfields and 3 Wolfdales, I've noticed the following:

1) not all cores are made equally: even on a single core Wolfdale, there can be 5C+ variance from core to core. That's just silicon wafer process variance. When you throw in 2 cores onto a single packaged chip, there's bound to be more deltas between the 2 dies.

2) Thermal diode sensor issues: this is real. 3/7 CPUs I've played with seemingly had the sensors stuck or out of wack (20C+ deltas between cores is just unreal). I think the most critical aspect is to gauge Trise between the cores and see how well the track with one another.

3) To a certain extent, lapping can compound the issues if you don't know what you're doing. After lapping 3 CPU's now, i can tell you this: Intel stock IHS sucks! Every single one i have lapped showed an issue with being concave in the middle. Coplanarity between HSF / IHS was well off as demonstrated when i took saw how the TIM application was not uniform. Lapping definitely helped in this regards and help temps to go down.

So to summarize, lapping shouldn't make things worse. I think Intel still needs to finalize their wafer process for these thermal diode issues since they transalte to customers having huge disparity in temps ( i expect more steppings soon as they ramp up the 45nm). FOr the time being, as long as Trise tracks with one another, I think you're okay.

 

[Syzygies]

Originally posted by: ST
Just to play devil's advocate here,...

I used to see a 4 C spread between my cores. After lapping I see an improvement across the board, but a 10 C spread between the cores. Yes, my Core 0 may run hot, period, but clearly my actions have some effect. Play with screw tensions, rotate the TRUE 180, reapply TIM, and decide to accept what I have, when I like a roll of the dice. I'm happy to trade in some of my great results on the other cores, to get Core 0 down.

This was my first lap. I wish I knew to go to an automotive store for sandpaper. The Q6600 was easy to lap while dry, but there's no reason not to lap the TRUE wet, with dishwashing soap as a lubricant. The TRUE was very tricky to lap; it tended to stutter, whatever I did. I may leave the Q6600 alone but relap the TRUE.

 

[imported_ST]

Originally posted by: Syzygies

Originally posted by: ST
Just to play devil's advocate here,...

I used to see a 4 C spread between my cores. After lapping I see an improvement across the board, but a 10 C spread between the cores. Yes, my Core 0 may run hot, period, but clearly my actions have some effect. Play with screw tensions, rotate the TRUE 180, reapply TIM, and decide to accept what I have, when I like a roll of the dice. I'm happy to trade in some of my great results on the other cores, to get Core 0 down.

This was my first lap. I wish I knew to go to an automotive store for sandpaper. The Q6600 was easy to lap while dry, but there's no reason not to lap the TRUE wet, with dishwashing soap as a lubricant. The TRUE was very tricky to lap; it tended to stutter, whatever I did. I may leave the Q6600 alone but relap the TRUE.

I had the same problems with lapping my Scythe Ninja. Once you've done it enough, you learn just to go slowly and that will lead to better results. Fortunately, I was able to double check my lapping since i had stock intel HSF as well as a Zalman unit to bench again. BTW> Automotive stores are rip-offs. Just go to Orchard Supply, much cheaper and you can just get single sheets.

 

[Syzygies]

Originally posted by: ST
Just go to Orchard Supply, much cheaper and you can just get single sheets.

Actually, I did go to OSH (Concord), they had a gap between 600 and 1500 grits. I filled in 800 and 1000 (which I may not use) at an auto supply store.

 

[imported_ST]

Originally posted by: Syzygies

Originally posted by: ST
Just go to Orchard Supply, much cheaper and you can just get single sheets.

Actually, I did go to OSH (Concord), they had a gap between 600 and 1500 grits. I filled in 800 and 1000 (which I may not use) at an auto supply store.

INteresting, down here in San Jose, I was able to pick up 300,600,100,1500, and 2000. A little bit of overkill, but oh well.

 

[Syzygies]

Here is an update with what I have been able to learn about Intel Quad Core numbering.

The following montage illustrates what I believe is most likely to be the correct answer:

Intel Quad Core numbering

Disregard the issue of "index origin" and pay attention only to the linear order: 0,1,2,3 is the same physical information as 1,2,3,4.

The upper left image is a composite, starting with Gigabyte's own image of their GA-EP35-DS3P motherboard. The board is oriented as it would be in a vertical case like my Antec P182.

The upper right image is taken from

INTEL QX9650 Core 2 Extreme Processor Review - A Closer Look

I cannot tell if this image comes directly from Intel, or if the text has been added by the author based on information from uncredited sources. I have written the author, but he has not answered.

The lower right image comes from my Intel Bad Axe II motherboard manual; it's the best illustration I know showing the orientation of a processor in a 775 socket, for confirming that my composite is correctly oriented.

The lower left image comes from Artic Silver's directions for applying AS5 to an Intel quad core processor.

A forum thread that agrees with the above article is

Core layout on a Quad ?

There are no sources given to convince anyone than the information here is not simply an internet rumor, repeating what one person once thought they believed. In the worst case, the author of the first article photoshopped his Intel image to more professionally present the information he learned here, without giving sources.

A contradictory article was once

Will you be overclocking a P45 mobo? READ THIS!

Before my exchange with the author in the comments, he had written

This is particularly noteworthy with quad-core processors because, if you?re finding core two and three drop off under Prime95 load, it?s down to the fact that the two CPU dies are not identical and while core one and four can hit the FSB you?ve set, cores two and three are having trouble

Pressed, he revealed that he had been told that indeed, cores 1 and 4 share a die, cores 2 and 3 share the other die. This flatly contradicts the other information.

Finally, I wrote Intel asking for technical support for my various quad core processors. I kid you not; I quote from their response:

We do not have that information available, however you might want to check the datasheets and design guides of this processor located at the following website:
Intel® Core?2 Quad Processor

For more information on this matter, you can also contact one of our authorized distributors and sign a Non Disclosure Agreement (NDA).

I could not find core numbering in any document at that site. I am astonished that this could be covered by an NDA, but that could explain everything. It's not that people writing in this business have a disregard for the virtue of giving references, it's that they were breaking their NDAs by saying what they did, and don't want to get busted on this?

Is my diagram right? Probably. Am I certain? Absolutely not. I cannot rule out a single source that actually didn't know for sure, reinforced by the echo chamber.

Anyone who thinks I need meds should reserve judgment until they've read through this NYTimes three part blog on a particular Crimean war photograph:

Which Came First? (Part Three): Can George, Lionel and Marmaduke Help Us Order the Fenton Photographs?

 

[Syzygies]

I finally have better evidence, by inference rather than by direct Intel documentation, that my guess above is correct:

In this article there is a labeling four core diagram:

INTEL QX9650 Core 2 Extreme Processor Review - A Closer Look
penryn_quadcore.jpg

In this article there is a two core diagram labeled in exactly the same style:

Intel's 45nm Dual-Core E8500: The Best Just Got Better
Penryn.png

I can readily find both images in unlabeled form at Intel's site, and I am perfectly capable of overlaying labels like this using Photoshop. Finding one labeled instance in an article, and I would worry about an overzealous writer 'shopping on labels so their article didn't look so hoaky, unconcerned with the authority that this lends information that could be a mere rumor.

Two labeled instances in identical styles, on different sites by different authors: Their source has to be the same, and is very likely to be Intel, either in public documents I haven't found, or in press kits I don't have.

 

[Drsignguy]

Well now the evidence is clear that the core info is read from left to right. Still, by the layout on the MoBo puts cores 3 and 2 towards top and by my calcs, those 2 are running cooler than 0 and 1. So, By placing the HS ( in my case the Tuniq 120) doesn't jive with the weight as All seems to be secure and at the same torque.

 

[aigomorla]

lol....

This to help you decide on how to apply TIM? because unless were talking about bowed waterblocks here, air sinks puts too much torque to give you perfectly even temps.

EDIT: yeah i was tripping about the core placement. :\ Havent seen these pics in such a long time. :T

 

[Syzygies]

Originally posted by: Drsignguy
...the layout on the MoBo puts cores 3 and 2 towards top and by my calcs, those 2 are running cooler than 0 and 1. So, By placing the HS ( in my case the Tuniq 120) doesn't jive with the weight as All seems to be secure and at the same torque.

Yes, my core 0, or cores 0 and 1, or sometimes 0 and 3, have always run hotter. If this has anything to do with the cooler to IHS contact, then I'm always seeing a contiguous range of "best contact": 1,2,3, or 2,3, or 1,2, respectively. But I find it hard to believe that my "best contact" patch is shifting. It could be.

Once one has secured a massive cooler, my intuition is that the screw torque more or less makes it a single object with the IHS. The weight issue is more a concern about damage elsewhere to the motherboard. E.g. without a backplate, authors have worried about the motherboard flex causing other components to lose contact with their heatsinks, crashing the board.

Nevertheless, I'm curious if changing screw tensions has any measurable effect. A core chart we believe gives me the confidence to try experiments.

As is oft discussed in other threads, the issues may be beyond our control: How is each die soldered to the IHS?

If I had any idea what scale to be experimenting with, I'd experiment with a shim placed beyond Core 0, between the cooler and IHS near the edge of their contact, to infinitesimally flex down on Cores 0 and 1 a bit. Tuning a car or anything else is the same story, one gets as far as one can with theory, then a bit further through inspired tweaking based on partial intuitions.

I've read that the particle size of Artic Silver 5 is about that of 1200 grit sandpaper scratches. I'd imagine that any modifications to lapped surfaces should be no more than several times this scale. What thickness is a line left by gold foil, the kind used for writing one's name in a book and so forth? The Intel thermal design PDFs all show scratching a groove in the IHS to take a thermal sensor; one could make a groove then fill it with a wire filament.

I'm waiting for the other shoe to drop. Realistically, I'll try relapping, better TIM application, playing with screw tensions, then go to water cooling if heat rather than voltage remains my overclocking bottleneck. Or let sleeping dogs lie, and start programming in parallel.

 

[Drsignguy]

It would be nice to see your results. Especially if it were + results!

 

[pituCaleya]

Hi everyone,
I was just looking for this information, since I got a Q6600 and saw the discrepancy between core temps. I was worried about heatsink + fan weight, (it's a Noctua NH-U12P, not as massive as a TRUE or a Copper Ninja, but quite heavy nonetheless), leaving part of the CPU with less heatsink "contact", so I did quite a foolish test, I put the computer upside-down, and also on its side, and the temps were still different between cores by the same margin. (My core #0 is 4-6 degrees C higher than the others).

So, I don't know, maybe this test was useless, but now I feel more confident about the heatsink attachment to the CPU, so I guess I'll think about lapping, although, if core 0 is hotter for so many people, maybe we've hit some hardware design issue, what do you think?.

Regards
Antonio.