Help with a lapped QX6700

[Big_Daddy]

Posted this over in the Intel section, but maybe somebody here can shed some light.

I'm running a QX6700 and was having terrible temps, They were uneven, and very high. There was a 10 degree difference between the cores. And yes, I tried several mounts, nothing could even out the temps, or lower them.

So I lapped my cpu. I think it turned out very well. The temps dropped 5 degrees by lapping, and the cores got much closer. There is now a 5 degree difference between cores 0-1 and 2-3. Under load it's 56-56-51-51 (respectively).

Did I leave too much material on cores 0-1? Could that be what's causing the differences?

Is this the correct orientation of the cores?

Courtesy of Redcorn

Thanks to everyone..

[cobra_kai]

If that is the final picture of your lapped cpu you definitely need to take some more off. When I lapped my first cpu I found this thread very helpful

http://forum.overclock3d.net/showthread.php?t=4735

It has suggested grit sizes and pictures for each stage

[Big_Daddy]

heh, no that's just one somebody doctored to show where the cores are, believe me. It's shiny copper. Mirror like. Down to 2000 grit.

[Bojamijams]

heh, no that's just one somebody doctored to show where the cores are, believe me. It's shiny copper. Mirror like. Down to 2000 grit.

Did you lap the base of your waterblock as well?

And I don't think the cores are split like that. The quads are two cores joined together (hence not a true quad, more like a 2+2) down the middle

[Big_Daddy]

lapping a d-tek v2 will destroy the perfect finish, and remove the bow. Neither of which I plan on doing. And yes, I know it's a 2+2. But depending on how you lap it, you can get uneven temps. (during lapping I got cores 0-2 the same and 1-3 the same, with a 5 degree difference between the sets)


0 - 50
1 - 55
2 - 50
3 - 55

[pur]

Well its possible that the IHS is just not connected 100% to the cores. It does happen you know...

[Wolf132]

qx6700 overclocked or stock at those temps?

[WoZZeR999]

The two hotter cores could just be... hotter too. Sort of like how two people can have <insert random dual core> and with the same settings they get different temps.

[Big_Daddy]

Stock, but at 100% load. OC'ing didn't raise the temp at all. (only went 2.8 I wanted to even the temps before going hog wild)

And I know that temps can be higher on some cores, but they did change on one of the laps. Notice Cores 0-2, 1-3 were the same. Now 0-1, 2-3.

Notice on this pick,



Maybe too much material on the left side.

[skinnee]

I don't know if you follow the RealTemp thread here on XS, but the DTS sensors on the cores that measure TJunction are not the finest calibrated sensors, and yes, the can be off by as much as 5ºC. The sensor deviation is the primary reason RT has the adjustments you can make to balance out your readings. I strongly recommend performing the calibration setup that unclewebb has documented.

Also, for the latest versions of RT in the thread, check towards the last few pages.

[cobra_kai]

It's really hard to say what the problem is without more tests. The two cores that are higher temp changed, but you also changed the experiment by lapping it. It's possible the first discrepancy was caused by a poor heat spreader, and the second was caused by deviations in the two dual cores glued together. The other possibility is that the different cores were processing different instructions. I know, you said that CPU is at 100%, but depending on the type of instruction running on each core (integer add, integer mult/div, floating point op, branch resolution, shift, etc) more transistors on that core might be lighting up causing higher/lower temps. Also, as skinnee pointed out the temp sensors are not the greatest. Since the temp difference is within the +/- for the sensor I probably wouldn't worry about it.

[Big_Daddy]

I was running Prime95 with small fft's. I relapped it, paying closer attention to the left side (cores 0 and 1) and now the temps are within 3 degrees. And before everyone states it was just a remount that fixed it. I mounted it 3 or 4 times each time. Varying the paste etc each time.

[skinnee]

yeah, +/-3 degrees is well within spec on the DTS sensors.

[gummybearx]

i lapped my qx6700 not because a temp spread but because it ran super hot, and the pattern the as5 left when i removed the block looked really scary. mine is all copper. at least take it off to the copper all around, then polish it up with > 1000 grit.

[smee]

Hey Big D, I've got a QX6700 (ES), I also lapped it, I got about 8c better temps under load (on all cores), but hardly any difference in Idle temps...

Before lapping, here are the averaged temps of all 4 cores (watercooled BTW):
36c Idle, 53c Load

After lapping:
35c Idle, 45c Load

Temps were with a CPU only loop with Dtec FuZion V2 w/out quad core insert (the block wasn't lapped either, completely stock from the factory).

I'm sure glad i lapped the processor.
Best wish's for ya!

[Big_Daddy]

nahh.. big fingers.

BTW, for those still wondering, this is what it looks like now.

[BlueAqua]

Holy shiot. Big Daddy that looks amazing. I think I'll lap my QX6700 and test temperatures with my AC Double Impact and the Swiftech GTZ. We'll see how that goes.

[philbrown23]

sometimes quads have 2 hotter cores and for some reason they are usually core 2-3 although 0-1 get used more.

[skinnee]

the lap job did go very well!

what did you use as a flat surface?

sometimes quads have 2 hotter cores and for some reason they are usually core 2-3 although 0-1 get used more.

Not always the case...any one of the cores could be the hot core. There is no pattern to a specific core, the DTS sensors are not calibrated or super accurate and can swing 5 degrees either way.

unclewebb has done more than enough research to prove this out.

[Big_Daddy]

peice of glass from Home Depot. Just went up and asked the guy if they had any scraps. Turned out he had a perfect peice. 8"x12" He taped the sides, and handed it to me. Gratis..

[skinnee]

Nice score!

[Waterlogged]

If that is the final picture of your lapped cpu you definitely need to take some more off. When I lapped my first cpu I found this thread very helpful

http://forum.overclock3d.net/showthread.php?t=4735

It has suggested grit sizes and pictures for each stage

I'm sorry to say that anything above 800 grit, is purely for bling and does not help thermal transfer one bit.

BD, flat is one thing, parallel is something totally different. If you have some calipers or micrometers or better yet, a height gage of some kind, those would be perfect for checking parallel.

[Big_Daddy]

I have calipers, but not in the range needed.

[cobra_kai]

I'm sorry to say that anything above 800 grit, is purely for bling and does not help thermal transfer one bit.

BD, flat is one thing, parallel is something totally different. If you have some calipers or micrometers or better yet, a height gage of some kind, those would be perfect for checking parallel.

Perhaps it is only for bling, but that is a big part of why a lot of people do big fancy WC modding projects. At least it is definitely a big reason for my interest

[skinnee]

Bling should always be secondary to performance.