7980XE (ES) very inconsistent core temps

[rsnubje]

Hey all,

Recently I've delidded a 7980XE ES and before AND after, the temperatures of the individual cores have very big differences between them.

For example core 1 can be 70c, while core 12 is 90c and this is the same for several other cores. Some are very cool, some are really hot.

I've seen it before with a previous 7980XE ES I've had, so I'm wondering, is this an ES thing, or can it also be seen on retail chips?

And can I do something about it? So far the chip has been relidded with a new copper IHS with Liquid metal, but also with the original IHS, but there's not really a difference. Letting the IHS loose without glueing it back with silicon, the temps are even worse.

[Leeghoofd]

Could it be your die is not flat Joost ?

[rsnubje]

That might be the case. What's a proper way to test that?

[Fasttrack]

Hi,

Although my problem is not so big as yours ( I had a serious issue with cores 2 and 12 ), the only solution I thought of was to go to bios, select per core adjustment and

lower the multiplier ( and voltage ) of these two cores by 1 step. Meaning, running 16 cores at 48X with 1.26V and the two "problematic" cores at 47X with 1.22V. This ironed

out the big fluctuations at a +/- 5c under load status.

[Shroudster]

2 hours ago, Fasttrack said:

Hi,

Although my problem is not so big as yours ( I had a serious issue with cores 2 and 12 ), the only solution I thought of was to go to bios, select per core adjustment and

lower the multiplier ( and voltage ) of these two cores by 1 step. Meaning, running 16 cores at 48X with 1.26V and the two "problematic" cores at 47X with 1.22V. This ironed

out the big fluctuations at a +/- 5c under load status.

The big question is though , is/was your chip also an ES or is it a retail sample?

Also checking in here since i'm the delid guy in question here. No abnormalities on the pcb/die and coverage/contact with the die/IHS was spot on.

So yeah been driving me crazy this week annoying chip this be

Only other thing that came to mind would be lapping the die a little but even for me that seems excessive and truly a last resort + owner of cpu must deem it neccesary.

[Fasttrack]

It is a retail chip.

I have also delidded my chip with Roman's tool.

As Firekiller ( Stavros ) - good man and personal friend - advised in another thread, lapping the die is useless unless you use direct on-die cooling.

Besides that it is pretty dangerous. I personally would never try doing something like that. I do not have the knowledge, nor the experience to play Russian roulette with a

2000 Euros cpu. 

[Splave]

Middle of die is high AF the stock IHS should accommodate this. I doubt the aftermarket IHS you purchased accounted for this. The intel IHS is concave inside to match the convex die. Some cores will also always be warmer than others as some are near warmer parts of the die IMC/FVR/ etc. I suppose removing the left over silicone on the pcb is not an option for you? def helps. It is es might as well lap the die even some 2000 grit for 50-100 strokes should be a benefit. 

[subaruwrc]

I think the liquid metal is not consistent enough with heat transfer.

try kryonaut or other paste if it solves that.Ive tested a similarly liquid metal relidded 7980xe retail and that was the symptoms too. hottest core is 105c with 90-85 the coldest under prime.

dark rock pro 4 cooler.

 

 

[Shroudster]

15 minutes ago, Splave said:

Middle of die is high AF the stock IHS should accommodate this. I doubt the aftermarket IHS you purchased accounted for this. The intel IHS is concave inside to match the convex die. Some cores will also always be warmer than others as some are near warmer parts of the die IMC/FVR/ etc. I suppose removing the left over silicone on the pcb is not an option for you? def helps. It is es might as well lap the die even some 2000 grit for 50-100 strokes should be a benefit. 

Aftermarket IHS is actually the better one temp wise. Atleast with that one it is able to run at the volts/clocks.

Ironicly the other 2 chips I fitted with the same IHS ran allot cooler so naturally I was checking this IHS if it were perhaps a production fluke. So far seems like that isn't the case.

Stock IHS+ liquid metal and stock IHS + MX2 (only stuff I had laying around here) both resultated in pre-emptive shutdowns due to temperatures being past 100C mid way in R15 with the same cooling setup.

From the start pcb and IHS were both completely cleaned since without oldkit the best temps are achieved   (naturally reseal with minimal kit to fixate it not raise it)

[Splave]

what cooling are you using ? and what vcore? 

[Shroudster]

29 minutes ago, Splave said:

what cooling are you using ? and what vcore? 

Current screen is from improvised test on a X61 max rpm/max pump speed at +- 25 ambient. For that screen R15 was run at 1,2Vcore.

Tommorow it will be back at rsnubje where a custom loop with a 480 rad in push/pull will be used.

[Splave]

wtf 1.2v hits 100c? something is broken

[Fasttrack]

48 minutes ago, Shroudster said:

Current screen is from improvised test on a X61 max rpm/max pump speed at +- 25 ambient. For that screen R15 was run at 1,2Vcore.

Tommorow it will be back at rsnubje where a custom loop with a 480 rad in push/pull will be used.

^^^^^^^^

I have to agree with Allen ( Splave ). Running mine at 4.7ghz with 1.19Vcore set in bios, 2.00VccIN, 3.0ghz uncore with 1.12V and Real Bench v2.43 looping for 1 hour, results in 67c max in the hottest core.

Something is definitely very wrong. 2 X EX420 rads and a double D5 ( EKWB ) pump here. 

LLC6

Edited August 9, 2018 by Fasttrack

[Fasttrack]

This is my daily setting.

I can go as high as 4.9 core, 3.2 uncore with 1.31V and 2.10VccIN with LLC7 for 2D and 5.0ghz with 1.325V for 3D and still I am below 85c in the hottest core.

[Shroudster]

Yeah obviously something is wrong with it.

Million dollar question is what exactly is the issue.

Temps are shit with both original IHS and full copper one.

[Fasttrack]

I am going to try and help ( if possible ).

Here is the most detailed procedure I FOLLOW when delidding . I always use der8auer's tools.

So,

1) Obviously the IHS is correctly removed.

2) Remove all factory black silicon from the perimeter of the pcb.

3) Clean everything very well with isopropyl alcohol. Wipe the die with soft cloth. Lap the inner part of the IHS with 320-800-1500 sand paper until the bronze appears and 

mirror shine it.

4) Apply a very small quantity of thermal Grizzly Kryonaut both on the die and IHS and apply evenly down to a very thin layer.

5) Apply new high temp black gasket silicon on the perimeter of the IHS.

6) Re-insert the cpu with the IHS correctly aligned on the pcb in the delid tool and use the curing pressure element of the tool to establish firm connection. DONT use extreme

pressure.

7) Let it cure for 24 hours.

If nothing else is wrong, install the cpu and use carefully a thin layer of Thermal Grizzly condactonaut on the IHS.

Apply your waterblock.

Normally you are done. 

[yosarianilives]

Really dumb question as you got the others correct, but when the IHS is on you can see the little RFID chip correct? If not the IHS is on wrong and you shouldn't even be able to post the system. I ran into that when I delidded my 7940x the first time.

[rsnubje]

Yup we were aware of that

[subaruwrc]

mine running 105C at 1.1v under dark rock pro 4 during some tests of prime in place fft lol .. 

[Shroudster]

12 hours ago, Fasttrack said:

I am going to try and help ( if possible ).

Here is the most detailed procedure I FOLLOW when delidding . I always use der8auer's tools.

So,

1) Obviously the IHS is correctly removed.

2) Remove all factory black silicon from the perimeter of the pcb.

3) Clean everything very well with isopropyl alcohol. Wipe the die with soft cloth. Lap the inner part of the IHS with 320-800-1500 sand paper until the bronze appears and 

mirror shine it.

4) Apply a very small quantity of thermal Grizzly Kryonaut both on the die and IHS and apply evenly down to a very thin layer.

5) Apply new high temp black gasket silicon on the perimeter of the IHS.

6) Re-insert the cpu with the IHS correctly aligned on the pcb in the delid tool and use the curing pressure element of the tool to establish firm connection. DONT use extreme

pressure.

7) Let it cure for 24 hours.

If nothing else is wrong, install the cpu and use carefully a thin layer of Thermal Grizzly condactonaut on the IHS.

Apply your waterblock.

Normally you are done. 

Apprecciate the write out.

Apart from the paste (liquid metal instead of standard tim), sanding on the inside of the IHS and the shorter cure time that is pretty much my procedure. (usually 3 hours for LGA1151 and letting it dry overnight for LGA2066)

Isn't the first delid either my guess would be this is the 15th or something for LGA2066 and the only guy that had wierd temps was pumping 1,35Vcore through an 18 core paired with an AIO (Usually 15-20 degrees drop is what they reach easy compared to pre-delid)

For reference the other delid with this copper IHS on a 7920X did about 65 degrees tops in prime with +- 1,3Vcore in a normal sized custom loop. (1x360 iirc)

So yeah currently most parts have been checked and double checked.

Hopefully we can get a nice imprint of the die itself to check if the shape is that much off the standard. Hell at this point it may just be a faulty tube of conductonaut considering i'm running out of other ideas of what it could be. 

 

Wel it seems an update is in order.:

After testing by rsnubje under the custom loop temps were actually good/okay ? (Based on 5x R15 loop, picture in OP was just a single run)

Pre-delid 4,7Ghz 1,25Vcore :

After delid 4,7Ghz 1,25Vcore:

Hell it can even do 4,8 on 1,3Vcore now :

Edited August 10, 2018 by Shroudster
extra's

[Splave]

much better!

[Fasttrack]

Devil's advocate here, lol

Of course a lot better. Still the system looks like it is not perfectly tuned.

4.8ghz CBR15 should give you 4700++ CB points.

Congrats though. Seems pretty much spot on.

[rsnubje]

It was on a SuperMicro C9X299 PG300 motherboard and I couldn't be arsed to finetune it for a simple score It was purely for reference purposes.

[Fasttrack]

COOL !

Target accomplished