Warning: Some i7-6820HKs and i7-6700HQ have Uneven Core Temps due to Uneven Heatsink

Is the issue bigger than just the tension arm on the new AW's? Find out here...


***News Flash!***

The proof that supports my original theory of the heat sink being uneven due to the thermal pads being too thick on the FET's above the CPU is true. You'll have to use a thinner thermal pad to allow the heat sink to seat properly.

Update (2-26-17): After analyzing numerous Heat Sinks, the culprit is in the design of the north part of the CPU plate. It bottoms out over the 'chokes' and causes the heat plate to sit at an angle which makes Core 0 and Core 2 temps much higher than the other two cores. Such a shame... It can be fixed, but it takes a lot of work to get it right if you want close to even core differentials.

There is another issue though in that the north side of the CPU plate with having only one screw, does not provide as much support as the south side that has two screws. A tri-mounting design is nothing new and it could have worked out more efficiently if the die was centered.

I've used some high end pressure paper...(This stuff is expensive.) Commercial grade stuff...

- And guess what? The side that is not sitting flush so happens to be on the side of the CPU where there is only one screw whereas on the other side there are two screws LOL. Coincidence? I don't know...let's ask a chimpanzee. Not rocket science.

So guys...there ya go...non-mystery solved. It's not the 6820HK...

The 15R3's that I've been working on also has a core differential of 30C! I have ran a bunch of stock benches to collect stock data, which I do on all machines that I work on before performing the repaste. I'll report back with the post repaste temps when finished.

Hope this helps those of you with the 15, sleep at night..

Bottom line. Let Dell to take care of this issue. As I've said before, the ONLY reason you would want to keep your unit is if you have a nice silicon and ONLY then would you want to give it the time of day to fix the issue on your own to keep that mobo. Otherwise you're just doing an injustice to the situation that is factually common and evident among many many 15 owners. ie.....They are listening and will take action in addressing these issues.

This type of QC issue should not exist in a $2000-$3000 machine. It's obviously the QC issue overseas and the poor workmanship that is being done there. Maybe Dell needs better management, an eagle eye on these workers and to switch up the thermal paste while at it.

Everything else is beautiful...the machine is great, but some of the things under the hood prevents this great machine to work like it's suppose to. But that's being addressed so hang in there.

Foundation is key and having good thermal paste along with a flush heat sink -> die to dissipate this heat is obviously important.

Solution for Customers: What should be done on Dell's part: A conditional type of a Recall of the AW15R3/AW17R4 and provide 1st Class service to those who have been affected. Fair is fair. ie... If a customer complains about their temps then take care of them right away.

Solution to fix the issue at the Factory: Perhaps make that single tension arm on the CPU side flat instead it being bend at a 90 degree angle near the screw. Check below, 9th picture down. Also, change out the thermal pads to a thinner one that sits over the FET's above the CPU. I've already tested this and it works well in conjunction with the bending of the tension arm.

Hope this helps...

It's obvious that it's cooking under there...look at that...


See those bubbles?! Air bubbles / pockets from the stock paste or any paste = NO GOOD! HEAT!


Laying down the pressure paper...


See how much darn pressure is on one side? Just crazy... (This is the CPU side.)


Evidence that the heat sink is uneven. Look at what side the pressure (red area) on the GPU is...yup... Also, the GPU has way more tolerance than the CPU although it is uneven like this, which is why the GPU temps aren't really an issue with a good repaste.


Plus look at them grooves on the CPU side of the heat sink! SMH ...This is why I lap the heat sink. Yes really...


***A Solution to fix the issue.***

On the CPU side of the heat sink plate there are three screws. One up top and two on the bottom. What you'll want to do is carefully bend the top tension arm upward so that when you tighten the screw in it'll create more pressure on that side of the heat sink plate, in hopes to even out the pressure of the heat sink against the CPU die. Also, use thinner thermal pads on the FET's above the CPU so that the heat sink can sit flush, thus providing a more even contact between the CPU die and HS.



WARNING: Avoid trying to loosen the two bottom screws to compensate as this is not the best option nor will it work as I've tested this already. All you'll be doing is creating a uniform GAP, which is not good for most thermal pastes due to its viscosity, let alone liquid metal! This is disaster waiting to happen. Pump out anyone? 100% pump out unsatisfaction guaranteed.

In some cases it may work, but if the pressure on one side of the CPU die is really bad and if there is no pressure on the opposite die of the CPU, it is counter intuitive because then you'd just be making an even space above the CPU die. This may work in the event of using a thick paste, however, having a gap between the CPU die and heat sink is not ideal for laptops.



Be careful not to bend it too much. -- (Here's the 9th picture. See the 90 degree turn at the tip of the tension arm? Instead of having it bent like that, perhaps make it straight so that when it is screwed into place, it'll provide more tension.



Thanks @hmscott for all your great posts. I'm also glad to be of help.

-

Although the purpose of this thread is NOT about REPASTING as a solution, the pictures below are for your viewing in hopes to satisfy any recommendations of respasting so that it is not mentioned. So please soak it all in.

Take a look at stock paste from my 17R4 lol...just crazy right? Spend all this money on R&D only to end up using crap paste? Yikes...Well numbers don't lie so I challenge anyone to put it to the test. The proof is in the pudding.

(These are from my repaste guide that just went live yesterday.)







---

On the bright side with a little work under the hood you can have the complete package.

***11 HOUR OCCT RUN on the 15R3***

EXTREME TEST!

Look at the temps! Max and Average.

• Liquid Metal Magic: Grizzly Conductonaut
• Thermal Pad Replacement: Fujipoly 14.0w m/k
• Realignment of the top tension screw on the CPU side of the heat sink. Click here.
  
• Lapped Heat Sink.
• Polished Everything Squeaky Clean.

6820HK Stress Test. PASS.


-

15R3 with STOCK TOOTHPASTE and STOCK CLOCKS. FAIL.


Thanks to @Mobius 1, @DeeX, @zergslayer69, @hmscott and everyone else that is contributing in providing valuable data!

 

Here is the repasting solution just to get it out of the way so we don't flood the rest of the thread with, "You should try and repaste.":

Okay so here's my take on the whole 6820HK issue. From my experience anytime the temps were off by a lot on some of the cores, it was almost always due to there not being enough paste on the die. The core order is kind of like the firing order on an engine so 0 is diagonal from 3 and 1 is diagonal from 2.

It could be that the heat sink is warped to where it is only making good contact across the die, diagonally, which is causing cores 0 and 3 or 1 and 2 or heck even 0 and 2 depending on which angle the uneven contact of the heat sink lies.

I would try repasting and using a healthy amount of paste to make sure that you have a nice thick even layer on there to ensure even coverage and to test the temps again. If the two cores are still miles apart from the other two, then it's likely a hardware issue. Also, just so we're on the same page, make sure that you're setting all the multipliers to the same number, ex. 36x on all four cores...

Another thing to try is to go into BIOS, set everything to defautl/factory stock, restart and enter BIOS again. -> Then make your changes on the OC profile or whatnot and reboot. Often times if you just set your settings without bringing it back to default, it can act very strange....BIOS is weird to begin with so it's good to make it a habit of always setting it back to default first, restart and to re-enter BIOS again on a fresh clean slate. Hope that makes sense..

Also.....I did notice something strange on the die of my 6820HK yesterday when repasting it. There seem to be a little bubble on the die that was underneath the surface? Let me see if I have a pic of it....but it could be possible that there are defected batches of these chips? Who knows....so far everything seems fine though temp wise. All my cores are pretty even and sometimes exactly even...

 

Toothpaste Hall of Fame:

Share your pics so we can let it be known!

 

AW17R4 - 6700HQ Stock.

Although we're focusing on the 6820HK, the over heating issue with AW is across the board even with the 6700HQ. (Mostly due to the thermal paste and uneven heat sink issue, but still...6700HQ or 6820HK, it doesn't matter.)

 

I had an issue like this with my gt80s before xotic PC shipped it, I had them paste with clu which made the issue exponentially worse ~20 C difference. I got them to build me a new laptop from scratch and this one still gets a 10 C difference but rubs on stock paste

 

Issues with uneven cores seem to be releated to heatsink issues making uneven contact. Part of the issue can be alleviated slightly by adding a bit more paste, but it does sound, to me, like hardware defect.

 

That would be a massive defect

 

That rules out the paste then? I had a feeling that it could be heat sink related...interesting...

Yes, Liquid Metal would make it worse with it being liquidy on a warped heat sink...

Correct. I always recommend being more generous with the paste with laptops due to the low pressure mounting vs a desktop where it has much more pressure so the amount of paste used isn't as important. However, with cases of uneven heat sinks and improper contact, compensating that gap by using thicker paste like ICD (which I'm not a fan of) wouldn't be the best option in my opinion. It'd be best to correct the underlying issue.

Yes it would. This is an issue with some of the Clevo's ...what a nightmare. - And sure enough the resellers use ICD to compensate for the warped heat sinks, gap etc....SMH....

 

It's entirely possible it could be related to the heatsink, if there is any warping. It was one possibility we considered looking at, but ended up switching systems as mentioned.

Well, it is used just because it is just a better paste than a good amount of other brands out there. (Not saying it is always the best, but just in general it has decent temp drops on just about any computer)

 

We all know repasting is often helpful when you get a laptop shipped with bad paste or a bad paste job.

That's not what we are talking about here.

We are talking about hot cores in a CPU from a pressure imbalance with the heatplate.

We have recently run across several 6820HK's that have at least 2 cores 10c-15c-20c hotter than the other 2 cores.

This is a heatplate pressure imbalance that re-pasting better or with better paste has no effect except to lower the overall temperature across all cores - the core temperature differential isn't affected unless you balance the heatplate pressure on the CPU evenly.

To warn new 6820HK owners with high core temperature differentials from re-pasting, and to instead return their laptop for another unit to get a better fitted heatplate.

We have a couple of new examples that can be used to help the new owner recognize the problem, and @iunlock is in the process of setting that up.

For now, please refrain from making re-paste recommendations for 6820HK owners with a couple of hot cores in this extreme differential range: 8c-10c as high as 20c, it won't help without also balancing the pressure applied to the heatplate over the CPU.

Edit: Edited to remove CPU failure possibility, as the uneven heatplate pressure imbalance has been found to be the issue causing the uneven core temperatures.

 

Internal failure as in defective hardware? And RMA is needed?

 

Yes, exactly.

I've had it happen with a couple of LGA CPU's - 980x and 2700K, and have seen it happen to a couple of BGA CPU's where the owner needed to swap for another one with the retailer to solve the problem.

XoticPC had one 6820HK hot core happen with mason2smart's new GT80S, and after several re-pastings XoticPC opened another GT80S - tested that the 6820HK was good - no high temperature core differentials - and shipped that to him.

We have noticed several new owners besides mason2smart, with an Alienware 15 and an MSI GT62 with 6820HK's that had 10c-15c-20c hot cores, which will need returning or RMA.

That's why we made the thread, so new 6820HK owners would know to look for this "hot core" problem. There seems to be more than the average number of 6820HK's with this problem.

Also, with the 6820HK being removed as an option in new builds, it made us wonder if this might be a batch problem with the 6820HK, limiting the number available for new builds.

 

Is this specific to this machine? I don't recall running into this same idiosyncrasy in the Clevo arena, not on the few machines that tout a 6820HK anyway.

 

It's unique to the particular sample of the 6820HK CPU, not an overall problem with a particular brand or model.

We have recently seen 3 6820HK's with hot cores, an MSI GT80S 980m SLI, an Alienware 15 2016 (pascal), and a MSI GT62VR 1070.

It could happen to any make / model with a 6820HK.

We are only suggesting that new owners check by doing a load test, with even load on all 4 cores, say at 36x multiplier on all 4 cores, and check what the temperature differential is between the cores.

Typically all CPU's will exhibit 0c-5c differential between 1 or 2 cores vs the others, the problems start when you have 5c or more difference.

Those "hot" cores lead the way toward thermal throttling. Reaching thermal throttling quicker than the other cores, and therefore limiting performance to the weakest / hottest core.

This problem with the "hot" cores in 6820HK can also happen to other brands, makes, models, it's a problem with the CPU if it has it, otherwise it's nothing to do with the maker.

Hopefully not too many of these will be found, so far 3 have shown up, 1 in a new Pascal Alienware 15, 1 in an MSI GT80S, and 1 in a new Pascal MSI GT62.

Check your core temps using a benchmark that evenly loads all 4 cores, set the core multipliers to the same setting, say 36x and see how much differential temperature is between cores.

0c-5c is normal, anything higher will interfere with OC'ing and even normal operation, with the "hot" cores thermal throttling before the other cores. 8c-20c differentials have been seen.

We also have a thread for this in the MSI section of the forums:
http://forum.notebookreview.com/threads/i7-6820hk-troubleshooting-high-temps-uneven-cores.797483/

 

I tested my 6820HK at 4.0GHz running AIDA64 for 1 hour. These are the results.
Is mi CPU OK?

 

I can't see the 1st image, but the 2nd image looks fine, just a couple of degree's C's off each other

1st image looks good too, only a degree or two apart.

Also, you don't need to run these tests for an hour. Start the test, reset hwinfo64 (what I suggest using) to start with fresh readings, and run for 5 minutes or so, just long enough to get a stable reading.

You are looking for core temps that are within 5c of each other, like 65c, 63c, 62c, 64c is a good spread.

If you have something like 65c, 75c, 64c, 76c, 65c - that's 10c+ off and those hotter cores will thermal throttle first, causing you to not be able to use all of your CPU performance.

Your's look good.

 

and this is my current cpu / gpu condition after repaste not sure what caused the issue, first the cable burn, now the center is like so messed up.. one after another I keep getting issues.. yesterday I had GPU throttling for first time, 920mhz for 1min, then gets back to 1126mhz, this happened after the latest Nvidia drivers, flashed to prema Vbios and older drivers with fresh install of OS, problem gone.. seems like something corrupted the VBIOS..

 

alienware is starting to build me a new aw 15, i wonder will they get rid of my old 6820 an replace it with a 6700. if so i want that 120hz screen.

 

Are you using the stock paste or have you repasted?
Is it the 17R4 or the 15R3 ?

 

I can't see either of them

 

Keep trying

I just tried again and got an error on the first one, saw the 2nd one, and tried again on the 1st one and it came up too.

 

Ran AIDA64 with the CPU and GPU going simultaneously, and the spread was 4-5 degrees. In just the CPU test, the difference was 1-3.

 

The CPU core differential was 4c-5c? That's not bad, not great, but not bad

When doing both CPU and GPU tests, I would imagine the load on the CPU would be greater - it's feeding GPU tests as well as crunching the CPU tests at the same time.

With previous testing I have found that the heavier the load, the more pronounced the core temperature differentials become. At idle, or low load, the differential isn't always as pronounced.

All CPU's have some differential, it's to be expected, it's when it gets up above 5c that it becomes a problem - limiting top end load - the hot cores getting thermal throttling first, holding back the cool cores.

Your results look good

 

I should correct myself slightly because the AIDA64 test included stressing the RAM on top of the CPU and GPU, so it really was an absolute worst-case scenario that 99.999% of users will never, ever see in daily computing. Maximum temp on any single core was 80 degrees, so even if there were more of a differential, it wouldn't even be close to throttling anyway.

Edit: And this is also on a Clevo system, which tends to feed too much voltage to the CPU anyway, so if I undervolt, there will almost certainly be even less of a difference.

 

Actually the differential stays the same under load, you will drop the overall temperature of all 4 cores with an undervolt, but the differential won't be changed.

 

Fair enough. Doesn't change the fact that there's not much of a difference and if there were, the system is nowhere near its thermal limit even under extreme load.

 

I think you are missing the point of this thread

It's the Core temperature differential, with the hottest cores leading the way toward thermal throttling.

If you have Cores that are 10c hotter than the others, you have lost a large amount of headroom for performance.

That's what we are talking about, not how cool a CPU runs in use, or how undervolting let's a CPU run cooler. All those things can be true, but if you have a couple of "hot" cores pushing temps 10c+ higher, you're going to be halted earlier on the path to full performance than if all your cores were at the same temperature.

Your CPU shows 5c hotter cores, which means those Cores are going to thermally throttle before the cooler cores.

If you can keep everything running cool enough to not thermally throttle at some specific setting, that doesn't mean you won't thermally throttle at higher multipliers - the point of the 6820HK is that it is unlocked and you can tune it for more performance than the default settings.

It's nice your current settings aren't thermally throttling, but now you can try tuning for higher performance, using your unlocked 6820HK potential fully.

 

I perfectly understand the point. I just don't care because my system is performing exactly as advertised.

 

That must be some pretty specific advertising copy

Just keep in mind that you can tune the 6820HK, and those hot cores will limit the top end potential, which is why we are watching for the hot core samples, so people can catch them and send them back.

It's not something to gloss over so quickly, as it will come back to take a bite out of performance later.

And, again, you are at the acceptable limit with +5c, so no worries

 

The days of my obsessing about overclocking have long since passed.

I'm not saying what's happening here isn't a concern. It's just not a concern for me.

 

You were concerned enough to test and verify, which is good, now you know

 

A lot of people are not very concerned now or they don't know. But later down the road, maybe they want to try overclocking for extending the life of the laptop. Then it is all too late bro It's bad enough with soldered i7... But even worse when it does not function as intended.

 

Just contributing a data point for the benefit of the group, nothing more.

 

Thanks!! +rep!

 

Hey all. i just got my 15 R3 6820/1070 Today and after a few Firestrike benchmarks i noticed That some of my core temps are 20c apart

Core #0 91c
Core #1 79c
Core#2 86c
Core#3 71c
Package 91c

Im kinda lost on what i should do here? I guess the only thing there is to do is contact dell and ask for a CPU replacement?.

Cheers.

 

Thats a faulty CPU. Repaste will help, but in your case its hardware.
Send dell a screeenshot and get a replacement.

 

Would you recommend getting a replacement Laptop or just making them come and replace CPU/Motherboard? and then if possible get them to use my thermal paste.

Thanks

 

They wont replace the CPU/motherboard for you. Even if they did why would you do that when you can get a fresh one.
20 C differential has to be replaced.
Repasting the new one will help.

 

Ah okay thanks It was just because ive been waiting for 3-4 weeks for it to arrive and dont want to wait again for a replacement but if that is the only option i guess that will do.

 

Even though this is a 6820HK centric thread, I can definetly say Intel's binning for mobile CPUs has been horrible at least since Haswell if not before.
My laptop has a 4710HQ, which I recently unlocked. And here's what it shows on Heavy Loads

While on lighter loads I get this:

A difference like this is most definetly not good and it's clear they are still doing it with "Higher tier" mobile CPUs

 

Yes, when you have a couple of good test runs with temperature readings showing a large core differential, with screen shots of a program like hwinfo64 showing the temperature readings you can submit with your RMA ticket.

hwinfo64 also has a line showing Thermal Throttling. That would also be good to show as part of the readings screenshot, to bolster your RMA case.

A high CPU usage benchmark that pushes the 2 hot cores into Thermal Throttling is what you are looking to submit. This will show the hot cores and show the thermal throttling caused by them.

Show in the RMA that you know there have been other 6820HK CPU's reported with this hot core problem, and you would like a new laptop with a new 6820HK CPU, as this hot core problem won't be solved with a re-paste.

Please come back and let us know how it works out

 

If you have an in home support agreement with Dell, you can request a replacement laptop or a replacement motherboard.

Normally I would suggest returning to the retailer asap so you don't get stuck with the bad CPU and have to work it out with an RMA to the maker, but if you have Dell support that might be your only practical choice.

When you contact Dell, let them know the CPU is bad and needs replacing, and let them provide you options.

In home replacement of the motherboard might be the quickest - make sure you verify it's a 6820HK motherboard and make sure to do a test run while the repair person is still there to make sure this new one doesn't have a problem.

Please let us know how it works out

 

replacement takes 7-10 days max

 

Thanks for the replies

I bought the laptop directly from dells website. So i will contact them and ask how they want to do it either give me a new one or get a repair man to come fix it.

What would you suggest tbh?

Just to make sure if i get the choice to get an repair person to come fix it what needs to be changed exatcly?

Thanks again.

 

You are reporting a bad 6820HK CPU, so the 6820HK CPU needs replacing.

The 6820HK CPU is solder to the motherboard, so Dell will either offer to replace the motherboard with the bad 6820HK, with a new motherboard with a new 6820HK.

Dell can do that with an in home repair visit - I think you need to request / suggest this - or via a cross ship RMA return of the whole laptop.

 

I'd bet if this ever goes to a full recall they would kick their buts for going BGA.

 

They'll chalk it up as just another cost of doing business just like with the GeForce 8600M GT back in the day, or more recently, Apple's recall of 2011 MacBook Pro models with the Radeon 67XX GPUs that would fail. They'll replace the parts the 3 years or whatever then move on to the next thing.

 

This is the kind of thing that could influence future methodologies. If the cost is high enough, and they think they can't mitigate the problem in the future, they'd have little choice but to change to replaceable CPU's to reduce costs overall

They do have rework stations to pull BGA parts and replace them, otherwise it wouldn't take too many complete loss motherboards with CPU and GPU to shoot the overall cost up high enough to make a design change using LGA look advantageous.

 

Are there any step-by-step instructions on how to best identify this problem?
Maybe I am just fighting with the idea that my device is affected, but I'd love to be certain.

In prime95 I get 98 - 86 - 99 - 87 °C as maximum temperature for the 4 cores.
Aida64 almost instantly signals that my CPU is throttling due to excess heat and shows 92 - 84 - 95 - 87 as the maximum temperature (I stopped after 32 seconds!).

 

That looks conclusive to me, you are getting 12c+ Temperature differential between Cores 0 / 2 and Thermal Throttling on those cores at stock default settings, that's the problem we are talking about.

It would help define the results better if you can reset the Max temperature between runs. The Core 0 temp Max shows high too in the 2nd screen shot.

Maybe use hwinfo64 to display temps, it has a reset button / Icon on the bottom left you can click to reset temps.

Also, let the CPU cool down to idle temps between tests, and then reset to get a full range of view from low to high temps.

After reset, screenshot the idle hwinfo64 view, run the test, and get a screenshot just as the test nears end - showing maximum readings, then let the CPU cool, reset temps, screenshot, run test, another screenshot just as test nears end.

I use Windows "snipping tool" to grab just the dialog box / area to screen shot - rather than the "whole screenshot", it's easier to see the focus of the shot this way.

What laptop is this? If it's Alienware, return for cause - the CPU has bad core temperature differentials - if you have in home service ask for a laptop or motherboard replacement to get a new CPU.

Or, you may need to return it for AW to ship another. Make sure you note the serial number of the laptop to make sure you are getting another unit in return

If it's not Alienware and you can return to the retailer - you are still in the "no questions asked" return period - you don't need to hassle with explanations - just tell them "its not a good fit for me", and then purchase again from another retailer - from another shipment batch to make sure you get another clear shot at a good 6820HK.

You can try explaining the problem to the retailer and ask for another unit, but that might be problematic trying to get them to understand the issue. You have to decide how to approach this return from the start, if it isn't going your way, try again with another service agent.

Please come back and let us know how it works out