Defeating thermal paste pump-out in high power GPUs

[EarthDog]

Looks like a normal application to me. Pump out is normal (regardless of the wattage).

I also think the results are part of the testing methods. I'm guessing the first results are from the card being 100% cold/room temp. That's the best it's ever going to be. You mention you 'notice' 90C hot spot temps, then run a test on a presumably warm/hot card (that takes less time to saturate). How long is your testing for each run?

 

[yoadknux]

Looks like a normal application to me. Pump out is normal (regardless of the wattage).

I also think the results are part of the testing methods. I'm guessing the first results are from the card being 100% cold/room temp. That's the best it's ever going to be. You mention you 'notice' 90C hot spot temps, then run a test on a presumably warm/hot card (that takes less time to saturate). How long is your testing for each run?

The testing method is exactly the same.

Cold Start -> GPU-Z -> TimeSpy Extreme Benchmark -> monitor temps.
Ambient is the same. Everything is the same.

The results are very consistent.

 

[Evilsizer]

personally looking at the paste spread, it doesn't look right to me. also the climbing temps over time, something is not right with that either. noticing you pictures, imo put the thermal pads on the IC's first. you have some looking like they are getting stuck/caught in the mounting hole for the screws. now if they are not factory thermal pads, then you have possibly changed the variables. if the pads have been changed and they do not have the same compression as the "new ones". then it seems like the newer ones are stiffer and preventing the correct pressure on the gpu die.

from what i see on the picture is that circle in the center of the die looks like a island. then past that is air in that gap till you hit some some paste after that. personally, think its a mounting issue. as much as i would like to get my hands on it to play with the mounting and look it over better. that is my 2c with what i see.

 

[EarthDog]

I see less of a problem with the thermal pads, lol. If they were too thick, the IC indentation on the pads would be more noticeable. As it stands, barely noticeable, bjt you do see the outlines so it's making good contact. I see what you're saying about covering the mounting, but it's not. The pad would be pinched and it isn't. Stiff thermal pads never prevented me from tightening screws down to their stops before.

Re: the pattern on the die, it's hard to tell. You don't get pump out if the gap is too large though. Perhaps the heatsink isn't flat? Check it. But nothing strikes me as off, still.

 

[yoadknux]

personally looking at the paste spread, it doesn't look right to me. also the climbing temps over time, something is not right with that either. noticing you pictures, imo put the thermal pads on the IC's first. you have some looking like they are getting stuck/caught in the mounting hole for the screws. now if they are not factory thermal pads, then you have possibly changed the variables. if the pads have been changed and they do not have the same compression as the "new ones". then it seems like the newer ones are stiffer and preventing the correct pressure on the gpu die.

from what i see on the picture is that circle in the center of the die looks like a island. then past that is air in that gap till you hit some some paste after that. personally, think its a mounting issue. as much as i would like to get my hands on it to play with the mounting and look it over better. that is my 2c with what i see.

I haven't touched the pads. Looking at other people's teardowns, their pads also "slip" next to the mounting hole, this is far from ideal, but I agree with EarthDog that if there was tension at this point, I'd see something like a slicing mark on the pad. Here are images from other teardowns for reference:




My current conclusion is that there's a mechanical mounting issue with ASUS 4090 cards. There's another guy online who reported high core/hotspot delta with his STRIX, until he changed the pads to putty.

https://www.reddit.com/r/nvidia/comments/116s1o9/asus_strix_4090_oc_repasteputty_mod_temps_now_way

I hope the CoolerMaster paste can hold out for longer... We'll see. Anyway I have PTM7950 on the way. It should resolve all issues. I don't really want to take the Putty route. It's pretty messy, and needs high quantities, meaning $$$. Replacing pads is also possible, but the stock pads have nice imprint, feel like it can only go downhill from here.

And final word... I run my card at 450W, no overclock, and even with "pumped" paste the hotspot doesn't exceed 85c. So it's not like my card doesn't perform as expected. Just wanna get to the bottom of this, curiosity.

 

[Evilsizer]

there is your key, the pads. if all he did was go from pads to the putty, then the pads do not have good enough compression when tightened down. this has nothing to do with thickness of the pads. you can get 1mm pads with different squishy to them, one might compress down to .5mm. when the coolers screw bottom out like they should. or it could only compress to .8mm which could cause the heatsink to not mount properly.

not sure if you have seen this video i posted in the past in another thread. you will see what about the squishy im talking about and yes it is a technical term

this is another video snarks did 2 days ago, just found it. im barely like 15mins in but you will a wide range of testing he has done of different setups. full putty, putty then copper shim then putty on that. some how the putty+copper+putty came out better temp wise on GDDR6x then just putty. looking at his page he has another video or two after this one i posted.

 

[EarthDog]

Good info!

For the OP, it's stock thermal pads. He only changed the thermal paste.

 

[Evilsizer]

Good info!

For the OP, it's stock thermal pads. He only changed the thermal paste.

but in that redit thread he did replace the thermal pads with thermal putty. it was able to squish down to where it should be been right from asus. that lead to better gpu core contact to the hsf, that ia way i said that is the key! down side of thermal putty is extremely hard to clean up, asus will deny your warranty since they do not use thermal putty. IMO high end gpus like that should really come with thermal putty from the factory on the ram/pwm sections. either that or they need to get the stuff in order on getting the right thermal pads. they could have been given the wrong thermal pads from the supplier. supplier being "they wont notice", we give them cheaper pad that do not match their specs so we make more money. there are to many things that could go wrong in the supply chain for asus.

also like for me personally if you watch the video ED of snark. that is the exact type of past contact i expect to see when i remove the sink. not what i see in the pictures the OP posted. that is my POV on the situation on that, you prolly have more hands on with different cards over time. i after all did go from 1060 6gb to a 2060 super that needed new past and pads. then not much longer after the 2060 super to a 3070 that has not been refreshed. i still like to watch what other people do out there. even at what people might call young age but not like i was when i joined back in the day. i just feel so much older, do not have the energy to be as active as i was. this overclocking thing was fun but time goes on cpu's overclock them selves i just lost interest. as long as the gpu temps and cpu temps look right when i get done, i am happy. i am prolly still one of few who see the PK3 as one of the best price/temp ratio thermal pastes. im not going extreme that one extra 1c for another 25% or 50% increase in cost is not worth it. LOL, i know i'm rambling now. i'm going to eat my feeling by the slice of pizza.

 

[EarthDog]

it was able to squish down to where it should be been right from asus.

I hear you. But if the screws lock down.... its getting what it's supposed to, at least in theory, right? A measly thermal pad will squish more easily than a heatsink. If it was the wrong thermal pad thickness I'd imagine the issue to be rampant (doesn't seem so at this time).

Also, the memory isn't really the issue though, right?

As far as what we see on the core die, it depends on the paste, really. Hiw the heatsink was removed, cold or hot, inconsistency in viscosity of the tim... so many variables. But I don't see bare die... my money is on adequate coverage.


We miss your thoughts here... but life does what it does. The door is always open!

 

[Evilsizer]

thermal pads do have different stiffness, maybe that is the way i should put it. we can still get screws to bottom out on other things when mounting them. it can still make bad contact for a number of reasons, it just has to be looked at. one thought is if the thermal pads are to stiff. even if you can still bottom out or fully screw the hs down. the stiff pads can cause the hs to bend in certain areas first and that can cause other issues. the metal relativity speaking is thin compared to the amount of pressure needed for mounting. that is why it is paramount manufactures get not only the thermal pad thickness right but also how stiff or soft they need to be.

OP if you have the card back together, run more high thermal loads. in GPUZ turn on your memory temp sensor, i wonder what those temps look like.

 

[yoadknux]

thermal pads do have different stiffness, maybe that is the way i should put it. we can still get screws to bottom out on other things when mounting them. it can still make bad contact for a number of reasons, it just has to be looked at. one thought is if the thermal pads are to stiff. even if you can still bottom out or fully screw the hs down. the stiff pads can cause the hs to bend in certain areas first and that can cause other issues. the metal relativity speaking is thin compared to the amount of pressure needed for mounting. that is why it is paramount manufactures get not only the thermal pad thickness right but also how stiff or soft they need to be.

OP if you have the card back together, run more high thermal loads. in GPUZ turn on your memory temp sensor, i wonder what those temps look like.

Memory temps are always good. 68-74c under max load, and unlike the hotspot, do not degrade.

 

[yoadknux]

More online info about ASUS problematic heatsinks? Coincidentally, I ran into an article published today in IgorsLab about a 4080 TUF with faulty thermals: https://www.igorslab.de/en/asus-gef...ot-ex-works-material-analysis-and-the-rescue/
In short, the factory paste application assumed a certain orientation of the core, which was different than what the card really was, and resulted in a 30c delta between core and hotspot. They were able to fix it by changing paste, but they didn't present any long-term results.

 

[Evilsizer]

is you gpu still running hot? what about the fans on there, are all 3 running?

 

[yoadknux]

is you gpu still running hot? what about the fans on there, are all 3 running?

Thanks for showing interest. I'll give an update.
The last time I repasted was on Thursday 04/04, using Cooler Master paste (see last post of previous page). Fresh paste gave me: 65.1c Core, 77.4c Hot Spot, 70c Memory.
Today (09/04), I ran the same benchmark under the exact same conditions, and got: 66.4c Core, 88.1c Hot Spot, 68c Memory.



So, in just 5 days after re-paste, Hot Spot went from 77.4c to 88.1c. And the delta (Hotspot Temp - Core Temp) went from 12.3c to 21.7c.

I went through 3 high quality pastes and all of them showed degredation after roughly the same time. The way forward:
a) Change from thermal paste to pad-based solution - I ordered PTM7950, but due to the state of my country (war...) delivery takes time - hopefully I'll get it by the end of the month
b) Change thermal pads to softer ones or putty - I like the memory temperatures so I don't want to change that
c) Increase mounting pressure by screwing the heatsink with washers - It's something people did on some AMD cards with lazy cooling, the problem is probability of cracking and squishing components - don't wanna take this risk on such an expensive card

 

[EarthDog]

Well, you either do "A" (b/c B and C I wouldn't) or you just enjoy your 100% performing, but warm running card.

 

[Woomack]

I feel like you are searching for problems that don't exist or are common while mounting any large graphics card cooler. Remounting often gives differences in temps. Sometimes even matters which screws are first. The cooler's surface is never perfectly flat, and you also never remove the cooler perfectly parallel to the PCB - always one side has some more pressure, so the stamp on the die is always different and doesn't represent the actual TIM spread on the die. On the other hand, if the cooler were perfectly flat, then under higher pressure, all of the TIM would be around the die. If you don't want to see it, then use less TIM.
All the photos with TIM spreading on the die look typical. At least, everything does not look much different than what I see after removing coolers from various new cards.

In high-wattage cards, the hot spot is typically a bigger problem than all other temps. The lowest hot spot temps suggest the best mounting.
Usually, thermal pad thickness causes too high hot spot temp - too thick and not enough cooler mounting pressure.
I don't remember, but the max hot spot temp without throttling on these cards is 110 or 115°C ... and you have under 90°C on a "stock" cooler.

Temps from post #34 are more than good. I don't know what do you want to achieve when you already run this card with fans at 30% speed, 500W and under temps for which other cards need much higher fan speed. You have temps like I have on my RTX4080 at 310W with a 3-slot, 3-fan cooler.

 

[yoadknux]

I feel like you are searching for problems that don't exist or are common while mounting any large graphics card cooler. Remounting often gives differences in temps. Sometimes even matters which screws are first. The cooler's surface is never perfectly flat, and you also never remove the cooler perfectly parallel to the PCB - always one side has some more pressure, so the stamp on the die is always different and doesn't represent the actual TIM spread on the die. On the other hand, if the cooler were perfectly flat, then under higher pressure, all of the TIM would be around the die. If you don't want to see it, then use less TIM.
All the photos with TIM spreading on the die look typical. At least, everything does not look much different than what I see after removing coolers from various new cards.

In high-wattage cards, the hot spot is typically a bigger problem than all other temps. The lowest hot spot temps suggest the best mounting.
Usually, thermal pad thickness causes too high hot spot temp - too thick and not enough cooler mounting pressure.
I don't remember, but the max hot spot temp without throttling on these cards is 110 or 115°C ... and you have under 90°C on a "stock" cooler.

Temps from post #34 are more than good. I don't know what do you want to achieve when you already run this card with fans at 30% speed, 500W and under temps for which other cards need much higher fan speed. You have temps like I have on my RTX4080 at 310W with a 3-slot, 3-fan cooler.

It's just curiosity. The issue I'm facing in this thread is real (the increase in hotspot), it just doesn't affect performance in any way as the core temperature is still very low.
I'm just curious and want to come up with "Ah, I did X and it improved something". Nowadays there's no point in overclocking a CPU and GPU, gotta find a way for that tinkering passion to manifest.

 

[Blaylock]

I know I'm late to this thread but I haven't seen this mentioned yet.

While it probably doesn't account for your spike in hot spot temps, what is the ambient temperature in the house? Does it fluctuate often? Here in the States we are approaching summer and the temperature is rising in the house. If you're running your tests on hot days you will get equally hot results.

Also I fully agree with Woomack. You are so far below the max temps for your card that this is really a non-issue and you are just obsessing. I know lower is better, and I agree, but at some point I would recommend just enjoying your card unless an actual issue arises.

 

[yoadknux]

I know I'm late to this thread but I haven't seen this mentioned yet.

While it probably doesn't account for your spike in hot spot temps, what is the ambient temperature in the house? Does it fluctuate often? Here in the States we are approaching summer and the temperature is rising in the house. If you're running your tests on hot days you will get equally hot results.

Also I fully agree with Woomack. You are so far below the max temps for your card that this is really a non-issue and you are just obsessing. I know lower is better, and I agree, but at some point I would recommend just enjoying your card unless an actual issue arises.

Ambient was more or less constant when I did those tests. Anyway absolute temperature is not what I'm after in those tests, rather delta between hotspot and core. I agree that the card is working as intended.

EarthDog, you mentioned previously that you have a 4090 as well, could you run the TS Extreme benchmark and monitor the Core and Hotspot in GPU-Z/HWInfo?

 

[EarthDog]

I'm on water. The difference between the hotspot and core varies from ~8-11C.