MSI GE66 i7-10750H - Thermal Throttling - High Core Temperature Difference - Performance Impact

orisos12b902b4

New member
Joined
Nov 17, 2021
Messages
3
How was the factory thermal paste application? Was it good? Was the thermal paste solid as rock or still viscous?
It was pretty much bone dry. Mind you, GPU temperatures were great regardless, so I definitely think it's a pressure issue on the CPU in particular. Either way, the new paste helped a great deal.
 

drek4

Active member
PRIVATE FIRST CLASS
Joined
Feb 18, 2018
Messages
821
It was pretty much bone dry. Mind you, GPU temperatures were great regardless, so I definitely think it's a pressure issue on the CPU in particular. Either way, the new paste helped a great deal.
Thank you for this valuable information, for now my GE66 runs nice and cool, but if I see any fluctuations, I will re paste shortly afterwards.
 

orisos12b902b4

New member
Joined
Nov 17, 2021
Messages
3
Thank you for this valuable information, for now my GE66 runs nice and cool, but if I see any fluctuations, I will re paste shortly afterwards.
You're welcome! Aside from a small issue here and there, they're very nice laptops. On that note, have you noticed any strange behavior from the speakers? One of the issues the client pointed out (which we've noticed a few times) is that sometimes the speakers "fade out" and get kind of quiet. They only use headphones primarily, and I believe an update for the realtek audio driver fixed it, but I was curious.
 

drek4

Active member
PRIVATE FIRST CLASS
Joined
Feb 18, 2018
Messages
821
You're welcome! Aside from a small issue here and there, they're very nice laptops. On that note, have you noticed any strange behavior from the speakers? One of the issues the client pointed out (which we've noticed a few times) is that sometimes the speakers "fade out" and get kind of quiet. They only use headphones primarily, and I believe an update for the realtek audio driver fixed it, but I was curious.
No issues whatsoever with mine, of course I mainly use it for gaming with headphones, but every time I turn it on with speakers and use it for browsing, I never had any issues.
 

TyraelArch

New member
Joined
Nov 14, 2021
Messages
14
Hello everyone. I think I solved the mystery. I repasted my notebook with Cooler Master Gel Maker, and ran some tests in Cinebench. The core temperature differences were reduced to around 13C, and the maximum CPU power raised to around 71W. However, 20 days later, most of the gains were lost, as shown in the image below shows:
Left graph = after repasting, smaller temperature differences than before with stock paste
Center graph = 20 days after repasting, core temperature difference increases again
Right graph = CPU power output, most gains lost, comparing right after repaste and 20 days later.
What happens is: some cores get too hot too fast, hitting 95C. The CPU thermal throttles to protect itself, even though other cores are colder and at around 70C. CPU power and performance therefore is limited.

After checking the videos below and understanding what this guy tested, I`ve come to the conclusion of the following:
1 - There is uneven contact pressure in my laptop, this explains why cores 0, 2 and 4 are always hotter. Since they are on the same side of the die, the other side has better contact, therefore lower temperatures.
2 - Repasting did diminish this effect, because there was a good material compensating this uneven pressure / contact. HOWEVER, since there is a mechanical, geometrical problem, either in the DIE or in the HEATSINK, this will accelerate the so called PUMP-OUT effect, and the thermal paste will begin to migrate to outside the contact regions between die/heatsink with the heat cycles during laptop operation. This is why after some days the gains were lost.
3 - This is most likely a manufacturing issue, the contact is not flush everywhere between die/heatsink, its uneven, the parts dimensions are not within necessary dimensions/form tolerance to create the ideal contact. This might also be caused by bad heatsink design, where geometry, screws positioning and quantity might not be ideal.
4 - This should be fixed by MSI R&D/Industrial Engineering team, however, the problem must already be known to the manufacturer and either they can`t fix it, or it is too expensive to fix it, considering mass manufacturing. Because the issue is not that apparent to normal customers and only enthusiast will notice, they might have decided to release the product as it is.

In these videos, the guy fixed the uneven contact by taking note of where the contact is bad (using the same idea dentists do when they check teeth contact when you bit, by placing a marker to see where the contact happens and where it is more intense, there are Fujifilms that can to this and are much better). Then he grinded and polished the heatsink to make the contact more flush. This resulted in INCREDIBLE gains in core temperature homogenity and CPU power output.

Pump-out effect in 1 minute:
 

Worfy

Member
Joined
Dec 7, 2021
Messages
49
Thermal Throtteling in stress tests like Aida64 , Prime95 , Cinebench is normal on values up to 94/95 degrees especialy if the laptop is at stock settings( no undervolt or underclock) . I whould be more worried if you have the same temperatures while gaming . Stress tests are ment to push all cores to there limits so yeah it could get up to 95 degrees on that ( here is the issue of how old the thermal compound is , some factory paste is crap and gets hard very fast) , but if you idle temperatures are under 50 on the CPU and GPU and while in load ( gaming) your temperatures hover around 80-85 degrees than its normal and ok.

Do not compare stress test results with normal ussage resaults , they are not the same thing. And if you want even cooler resoult while in normal usage such as gaming or editing or what ever , you can apply a undervolt up to -80mlv ( i found that more than -80 systems are not stable and crash allot) , and lower your PL1 and PL2 limits a bit. Allsaw in MSI Bios under Thermal Control > CPU> there is a option there caleed TCC offset what is set to 5 by factory. Setting it to 10 will force the system to get cooler a bit.

TCC offset is the thermal throttle limit what is set to 95 by default , seeting it to 90 lowers the CPU power intake thus lowering the temperatures.
 

TyraelArch

New member
Joined
Nov 14, 2021
Messages
14
Thermal Throtteling in stress tests like Aida64 , Prime95 , Cinebench is normal on values up to 94/95 degrees especialy if the laptop is at stock settings( no undervolt or underclock) . I whould be more worried if you have the same temperatures while gaming . Stress tests are ment to push all cores to there limits so yeah it could get up to 95 degrees on that ( here is the issue of how old the thermal compound is , some factory paste is crap and gets hard very fast) , but if you idle temperatures are under 50 on the CPU and GPU and while in load ( gaming) your temperatures hover around 80-85 degrees than its normal and ok.

Do not compare stress test results with normal ussage resaults , they are not the same thing. And if you want even cooler resoult while in normal usage such as gaming or editing or what ever , you can apply a undervolt up to -80mlv ( i found that more than -80 systems are not stable and crash allot) , and lower your PL1 and PL2 limits a bit. Allsaw in MSI Bios under Thermal Control > CPU> there is a option there caleed TCC offset what is set to 5 by factory. Setting it to 10 will force the system to get cooler a bit.

TCC offset is the thermal throttle limit what is set to 95 by default , seeting it to 90 lowers the CPU power intake thus lowering the temperatures.
In my laptop, the thermal throttling and hitting 95C also happens in games as well, along with drop in CPU frequencies, therefore resulting in lower peformance. Benchmarks and stress tests are used as a reliable way to compare and study.
Undervolt is already being used.

"TCC offset is the thermal throttle limit what is set to 95 by default , seeting it to 90 lowers the CPU power intake thus lowering the temperatures."
You are telling me to further cap my performance by making the thermal throttle to happen earlier. I do not care if it is 95C running hot, the issue is that the CPU is reducing frequency/performance to maintain 95C and not overheat.

The core issue here is: there is a huge core temperature difference, over 25C. This is not normal, repaste did help for a few days, but soon returned to original condition of HUGE core temperature differences.

This is caused due to improper heatsink/die contact-pressure. This is a manufacturing issue, you have either a performance cap in your laptop or increased laptop noise due to fans spinning at higher speeds to compensate the bad contact and heat transfer.
 

Worfy

Member
Joined
Dec 7, 2021
Messages
49
I curious dough how did you repaste it? Did you spread the paste by hand on the die by any chance? Becouse what you describe there can only be a shitty replacement job , no offence to you. Allsaw by doing that tou can spread it uneven making gaps in the compund. What you describe there of a concave or convex situation where the die dosent apply the same ammount of presure to the heatsink its basicly micromillimeter stuff what you cannot see by eye , and can only be measured by machines or a good micrometer , in the case that you did not actualy bend the heatsink when removing it.

If replacement of the thermal compund is done right the temperatures are not allowed to return to what they where especialy after 1 week of use. (did you whipe the thermal paste pads from the vram and mosfets and replaced it with normal thermal paste by any chance)? thats a white or pink thermal compund usualy more jelatinous that the normal thermal paste for the CPU and GPU.

For a even repaste you need a straight THIN line on the CPU die and a small pie sized blob on the GPU die straight in the middle of it . No more , no less.

My advice to you is do the repaste again , and im hoping to god you didnt whipe the pink thermal pads off ( that is special k5-pro) thermal pad not paste.

Allsaw by lowering your TCC offset to -10 you dont throttle faster , your temperature actualy gets to 90 a bit harder. And with a undervolt if the paste job is done right you shouldent even hit 90.


Mine after repaste for reference.

1. In game https://ibb.co/NmZMkRB You can se frequency and temperatures
2. HwInfo https://ibb.co/bgRsRny Detailed info
 
Last edited:

TyraelArch

New member
Joined
Nov 14, 2021
Messages
14
My journey is coming to an end, for now.

My MSI laptop has a manufacturing defect on heatsink/die, causing improper contact and pressure, lowering heat transfer.

I used a black marker on die and did several assemblies to check where the contact was good, then I sanded these places to make the rest come into contact, a flush mate between surfaces.

Before lapping: small contact area, manufacturing defect

Sanding the heatsink:

Improved contact, although not perfect:

Data and graphs:
First: Core temperatures after repasting with Cooler Master Gel Maker (some difference, but not huge)
Second: Core temperatures 20 days later, massive increase in core temperature differences, pumpout effect due to heatsink/die improper contact.
Third: After sanding the heatsink to improve contact with die.
Fourth: Power output during cinebench r23, yellow is after lapping/sanding the heatsink, managed to go from 56W (red) to around 76W, this means 37% increase in power output = more performancee!


There is still some uneveness in contact, next time i will make it a little better.

MSI, PLEASE FIX YOUR MANUFACTURING PROCESS!
 
Last edited:

TyraelArch

New member
Joined
Nov 14, 2021
Messages
14
I curious dough how did you repaste it? Did you spread the paste by hand on the die by any chance? Becouse what you describe there can only be a shitty replacement job , no offence to you. Allsaw by doing that tou can spread it uneven making gaps in the compund. What you describe there of a concave or convex situation where the die dosent apply the same ammount of presure to the heatsink its basicly micromillimeter stuff what you cannot see by eye , and can only be measured by machines or a good micrometer , in the case that you did not actualy bend the heatsink when removing it.

If replacement of the thermal compund is done right the temperatures are not allowed to return to what they where especialy after 1 week of use. (did you whipe the thermal paste pads from the vram and mosfets and replaced it with normal thermal paste by any chance)? thats a white or pink thermal compund usualy more jelatinous that the normal thermal paste for the CPU and GPU.

For a even repaste you need a straight THIN line on the CPU die and a small pie sized blob on the GPU die straight in the middle of it . No more , no less.

My advice to you is do the repaste again , and im hoping to god you didnt whipe the pink thermal pads off ( that is special k5-pro) thermal pad not paste.

Allsaw by lowering your TCC offset to -10 you dont throttle faster , your temperature actualy gets to 90 a bit harder. And with a undervolt if the paste job is done right you shouldent even hit 90.


Mine after repaste for reference.

1. In game https://ibb.co/NmZMkRB You can se frequency and temperatures
2. HwInfo https://ibb.co/bgRsRny Detailed info
I removed and reapplied K5 pro on all the memories and other components that originally had it, do not worry.
The degradation is due to uneveness of heatsink/die, accelerating the pump-out effect.
 

Worfy

Member
Joined
Dec 7, 2021
Messages
49
Well if your certain thats the cose then good luck fixing it.

Pump out effect does not ocur in 2 week btw it takes allot of time , coper is still a metal and does not change its form that fast. its not a plastic.

I bet when you reopened it the thermal paste was still on the die and some of it around it ,but thats due to the mounting presure not the pump out effect.Mounting preasure witch sits around 10-13 pounds depends on the vendor . That preasure for such a small surface area is preaty big. What you are showing there is micromillimeter measurments , and listening to 1 youtube rewiever does not mean that its the right choice , and i can ashure you 100% its not , based on the facts and examples listed bellow.

I hope you get it resolved , but if you had warranty on the machine is gone now.

''Fourth: Power output during cinebench r23, yellow is after lapping/sanding the heatsink, managed to go from 56W (red) to around 76W, this means 37% increase in power output = more performancee! '' 10750H is not designed to hold 76W of power sustained. its not a HK cip its a H (non overclockable) . You basicly screwed your warranty based on a hunch.

Btw Some cores heat up more than otheres due to the fact that they take the load of the next core to it to pump more performance as seen in this photo https://ibb.co/wg92DXS. (i7-10870h) I have TCC offset at-10 that why you see throtteling ocur at 90 and not 95. That's why you see core x and y throtteling and the rest stay cooler , thats normal CPU usage . as for the power limits , you have your PL2 short burst up to a ''X'' ammount of wats normaly 75-90W for a few seconds (remember thats PL2) (1 or 2 seconds) than the PL1 kicks in for sustained frequency lowering the power ussage drasticly , basicly around 40-50 w of power even lower if heavy throtteling occurs can go as low as 30W if needed , not more. The 10750h is not designed to sustain 4 Ghz on 6 cores , it holds 3.5/3,6 maby 3,7 if the silicon is good , its the lower end of the 10'th gen i7 line and its not designed to hold 60+wats of power for sustained time no mather how much you want.

I can imagine that in full throttle your hitting somwhere in the ballpark of 3.2- 3.3 Ghz on all cores , what you should have done is -70-80mlv undervolt , then lower PL1 and PL2 limits , and underclock the cpu each core by 500 mghz. In a cinebench score after you whould have had around sustained 3,6 ghz while temperatures whould have been under 90 after repaste alone. Better sed you whould have gained sustained performance by lowering the clocks and temperatures insted of 3.2/3.3 you whould have had around 3.6 sustained. Limiting power to the cpu is not the same as limiting frequency. Cant remember exactcly the core clocks of each clock but i think its somthing like Core 1 (5ghz) lower to 4.5 , Core 2 (4.9 ghz) lower to 4.4, Core 3 (4.7 Ghz) lower to 4.2 , Core 4 (4.6ghz) lower to 4.1 , Core 5 (4.5ghz) lower to 4.0 and Core 6 (4.4 ghz) lower to 3.9 or lower if needed . Sustained performance whould have been higher than what you where getting by throtteling. In any case you cannot reach 4+ ghz on all core sustained on that CPU its not possible whitout water cooling , and the CPU cannot hold more than 40/50w of power sustained for long periods becouse its not designed that way. What you see there jumps from 40 to 70 and goes back and forth is the PL2 turbo and PL1 sustained doing jumping jacks to keep the CPU as higher performance as it can.

A other way to save your warranty MABY is to buy a original heatsink part and replace it and hope to GOD they dont see any difference if it ever gets sent for repair.

My adivce for you is to stop for now do a bit of real reserch on the CPU you have and then take a decision if you are 100% shure that is the problem.

You can see here '' https://cpu.userbenchmark.com/SpeedTest/1053158/IntelR-CoreTM-i7-10750H-CPU---260GHz '' the 10750H turbo (avg) score based on a few results and silicon lottery. Max turbo frequency is 4.25ghz , so if you wanted more performance than that you should of done your reserch on the 10'th gen lineup.

10750H specs - https://ibb.co/qjx4LjG key word max TDP at 45w (PL1 limit) so if you wanted higher power output on the cpu your out of luck sir it cant be done.

EDIT : sorry for the bad drawing but take it as a rough estimate https://ibb.co/k8mqxY9

The COPPER heatsink is a soft metal and dosent mather on how much you grind and polish it it will never be a perfect flat surface , under a microscope there will allways be groves or small dents and will never make perfect contact with the die , to most that can be done is what the Quallity Control of a certain vendor beeit MSI , ASUS , Lenovo , whatever deems acceptable in there measurments. What you did there is basicly create even deeper groves in the coper and making it more uneven than it was actualy measured and passed by Q.C.

That marker test :)) is real bad and this is why.
1. the viscosity of that marker is far lower than that of even the lowest of thermal paste's out there
2. That ink can basicly just go in the groves or dents and not the whole surface area . (its like basicly a drop of rain in a whole) the outside remains dry.
3. You whould have been better of making a imprint whit a aluminium foil ( battery cable out of the socket and hold the power button for 30 seconds to drain evrey bit of leftover energy in the laptop before doing that to avoid any unwated shorts) and even then im 100% shure that you whould of not seen any difference what so ever.
4. Even a piece of putty , playdough whould have done the trick far better than a marker test.

Thats the role of the thermal paste to get in those groves and fill the voids so that the heat dissipation can be as even as possible. And thats why its allways requiered not to put to much thermal paste becouse where talking sub millimeter measurments what the naked eye cant see or the fingers cant feel. To much paste the thermal conductivity does not reach 100% the heatsink and does not dicipate it at 100% efficiency , to little its not enough to cover the surface area of the die some areas beeing more hot than others as a result. TjMax is set at 100 degrees celsius and the failsafe of the CPU shuts down the laptop at 115 degrees so that it dosent fry itself to death.

I urge you please to listen to this guy and hear what he says about power usage and thermal performance in laptops.

He talks about DELL laptop but its basicly the same for all laptops , becouse evrey thermal engenier of any company follows more or less the same principal.

1.
2.
 
Last edited:

TyraelArch

New member
Joined
Nov 14, 2021
Messages
14
Hi Worfy, thanks for your comment.

Yes, I am pretty sure the issue is caused by uneven contact and contact pressure in my laptop.
You can check this thread for more information on the subject: http://forum.notebookreview.com/threads/ic-diamond-24-giveaway-reliability-survey.584682/page-50
The ideal test would be using Fujifilm Prescale contact pressure films. However, they are expensive and hard to find.
The black marker works pretty well to identify regions of contact and I used it to adjust the heatsink. By the way, warranty is not an issue to me, however you should not try to adjust your heatsink block if you care about your laptop warranty.

Here is what a Fujifilm Prescale film look like by detecting high contact pressure areas (red) and low or no contact areas (white):
As you can see, there is only a fraction of the die concentrating all the contact.

Ideally, all the area where the die is would be pink, indicating a flush mate between heatsink and die, with homogeneous contact pressure. This is what an optimal contact for heat transfer would look like.


Pumpout effect happens because of thermal cycling, the copper expands more than the cpu die, and pushes the thermal paste out, this happens for each Hot/Cold cycle. This is caused by thermal expansion and heat.
Indeed it should not happen so fast, 20 days in my case, but I believe the uneven contact pressure accelerates this issue.

The i7-10750h has many different power level settings and is adjustable for laptop manufacturers and also by end-users. If you start to read about the overclocking scene and professionals, you will see that pushing higher powers is possible and safe. This might be done via delidding in desktop cpus, or using better thermal pastes, liquid N2, or even by lapping (sanding/grinding) the heatsink/die.
To achieve 80W sustained on a 10750h for sure is possible.

You can check this link, and scroll down to Cinebench R23 results:

As you can see, different laptop manufacturers, different models, different power level powers and benchmark results. This is due to laptop thermal constraints, heatsink size, fans, thermal interface material AND contact pressure!


The heatsink does not need to be perfectly flat, it would be ideal, but it is not necessary, neither cheap to achieve (mirror like finish).
I used a sandpaper with 140 grain size to adjust my heatsink block. As you can see on the graph below, using a grain size above 80 (higher is finer) yield not much difference.

I already saw these Dell engineer videos, thanks for sharing them.


You told me to do my research, trust me, I did. And the conclusion based on the tests I did, both phisically, by checking contact using a marker, and via stress tests and core temperature differences, shows me that I am in the right path. But please, lets discuss other possibilities and solutions for the issue I had initially (huge core temperature differences causing thermal throttle, both with stock thermal paste and even 20 days after a new repaste).
I believe the discussion is going to be great and very educational.

Best regards,
Alexandre.
 

Worfy

Member
Joined
Dec 7, 2021
Messages
49
I know about deleding , lapping even sub 0 temperatures if must be :) but not on a H chip , HK yea can be done and even there its a bit .....meh due to the thermal constraints of the laptop but H cips are locked force overclocking the power limits can only bring its early death in my opinion . As for lapping at least on a desktop i saw many lapp the actual die and not the heatsink (alldough very dangerous in my opinion) As for the Thermal throttle on those cores alone it can be a CPU issue (as in silicon) not necesseraly the heatsink.
I only sed do your reaserch becose once you go this route it aint no turrning back , i know that from past not so good experiences :) . And i mainly reserve deledin , lapping , even undervolting as a last option in case anything else fails.

What you could try to do is try to smothen the heatsink as much as you can out with that 140 grain then try and force pollish it but whitout any cream or substance just by a electric screwdriver and a soft brush , until it recieves as much as a mirror finish as it can. Problem is you cant sand to deep becouse you then run into much worse issues.

https://www.overclockersclub.com/guides/lapping/ you could allsaw try this but thats allot of sand paper and time :))
 

TyraelArch

New member
Joined
Nov 14, 2021
Messages
14
Check these results out. Got the time to run some tests (all at 20C ambient temperature).


Following sequence:
Cinebench R23 - 10 Minute Run
Timespy 3D Mark
OCCT - 10 Minute Run GPU Stress Test
Cyberpunk 2077 Mission
BF2042 Breakthrough 128 Players Match

First Graph:
Core temperature differences during Cinebench R23. This is where the heatsink adjustment shows its gains. Core temperature difference went from 25C to only 7C, THIS IS INSANE IMPROVEMENT!!!!
With this the score is so much better, I am hitting 8300 points multicore, before I was hitting 6900.

Timespy Results:
10700 GPU
7210 CPU
10042 Total

CPU and GPU temperatures, Frequencies, Powers:
As you can see GPU Temp max was around 77 in cyberpunk, which was something I was also checking.
In BF 2042 GPU was running at 65C and CPU was running at around 80C (PUSHING 4300MHz CONSTANT!!!!!)



MSI, please improve your heatsink mounting system and tolerances!
 
Last edited:

Worfy

Member
Joined
Dec 7, 2021
Messages
49
well looks like you did it :). Congrats . Good thing warranty is not a issue for you :)
 
Top