News Nvidia GPU partners reportedly cheap out on thermal paste, causing 100C hotspot temperatures — cheap paste allegedly degrades in a few months

[The Historical Fidelity]

If they didn't put so much useless bling on the cards (RGB crap for example and associated electronics) they could have probably afforded higher quality thermal paste.

Absolutely True and I personally would prefer build quality over unicorn vomit, but maybe not including those unnecessary things would result in customers passing them up for brands with more bling. Never underestimate the power of idiots with money lol. built like trash but lights up like a Christmas tree = take my money haha

 

[MichaelSchumacher]

Yeah I will also wait for Gamer's Nexus to do a thorough deep dive on this, hopefully with a good sample set from different manufacturers.

One key thing I am wondering about: if most (or all?) AIB are guilty of cheaping out on thermal paste, and effectively scamming customers with short-lived good performance that degrades, are the Nvidia branded Founder's Edition cards also likewise guilty? I would imagine Nvidia would specify a good thermal paste which would at least remain consistent through the life of the product.

 

[umeng2002_2]

Amazing how if it degrades in a period "of months" and we are 20 months after the 4080's release and this hasn't been widely reported since then, I'd say it's become an issue relatively recently. Personally I wouldn't buy any non Super RTX 4000 series card if only to ensure the power connector is 12V-2x6 and not 12VHP.

Because most people don't keep track of their hotspot temps. I repasted mine because I noticed the fan spinning louder than when I first got it. Low and behold, the hotspot temp was in the upper 90s.

 

[Geeee123]

Not something new. GDDR6X overheating was a problem from day one. My 3080ti needed full re-pasting of the die and a new set (with the proper thickness, thanks MSI) of thermal pads. This was the top-tier air cooled model and they chose to save 50 cents.

 

[gdmaclew]

Please change "looses" to "loses" in the sub-title.

 

[herrman976]

This thing seems to be older than one could think. Some years ago I ran into not one, but 2 used Gigabyte GTX10xx with detached chip due to overheating. Now I can better understand it...

 

[aberkae]

Vendors it's more cost effective to save a dollar on thermal paste then RMA the graphics cards that cost upwards of 2k usd. 🤪.
Luckily I have 3d mark score from 2022 and will rerun the test to see if anything was affected on my 4090 suprim liquid. The card barely reaches 60 degrees on 3ghz and +200 hz on the vram ( 450 watt) and often hovers around low to mid 50s in rt titles.
I suggest everyone rerun a previous benchmark and compare the results. The key is significant performance loss of 5% or greater. Things to consider; In the summer time ambient room temperature is definitely often as high as 10 to 15 degrees warmer ( in Fahrenheit) then in November when I purchased it so that might affect the results. 69*f vs 83*f now. Air conditioning will improve this to 75ish but might still affect results. I'll post my findings after work.

 

[awake283]

I dont understand the logic. Surely the amount of increased RMA's costs more than quality thermal paste right?

 

[awake283]

You can't just replace your Thermal paste. It's a lot more complicated. If you take it apart, you will need to replace all the thermal pads. Reuse is a bad idea. What thickness pads do you need ? Who knows? Better not get that wrong.
Kind of irresponsible to recommend folks tear apart their cards without more info on how complicated it can be..

For experienced people it's not too bad but that isnt the point. Your average PC user is not going to be comfortable taking apart their $2000 GPU.

 

[spongiemaster]

So I read through the article. No where do I see an actual reference to a cost or savings from the paste they are using. AIB's certainly aren't going to have months to test this paste for degradation before releasing their cards. 3 years is a pretty standard warranty. Why would they knowingly use a paste that degrades far before the warranty period ends? Based on reported margins for these cards, it wouldn't take very many replacements to eat into profits.

How do we know this isn't just the paste manufacturer screwing up? Either intentionally or through incompetence?

 

[Alvar "Miles" Udell]

Does make me glad I stuc

Because most people don't keep track of their hotspot temps. I repasted mine because I noticed the fan spinning louder than when I first got it. Low and behold, the hotspot temp was in the upper 90s.

I find it difficult to believe in this age of social media that this is as widespread of an issue as the article makes it out to be and is not limited to just certain batches or models. Given how even the mainstream aftermarket well regarded TIMs can degrade even when stored correctly quite easily, personally I've had tubes of IC Diamond, Arctic Silver 5 and MX-4 go out on me in as little as a year, it's entirely possible it's limited to just one batch of TIM, especially since QC the last few years has been slipping (remember the GPUs with their thermal pads with protective film still on them?), and bad TIM can look, feel, smell, and perform just like good TIM long enough to pass factory QC testing periods.

 

[KyaraM]

Does make me glad I stuc


I find it difficult to believe in this age of social media that this is as widespread of an issue as the article makes it out to be and is not limited to just certain batches or models. Given how even the mainstream aftermarket well regarded TIMs can degrade even when stored correctly quite easily, personally I've had tubes of IC Diamond, Arctic Silver 5 and MX-4 go out on me in as little as a year, it's entirely possible it's limited to just one batch of TIM, especially since QC the last few years has been slipping (remember the GPUs with their thermal pads with protective film still on them?), and bad TIM can look, feel, smell, and perform just like good TIM long enough to pass factory QC testing periods.

I tested my card all evening tonight. Bought the card 1 1/2 years ago, so if it had bad TIM like in the article it would have shown by now. The card in question is a Gainward Phoenix RTX 4070Ti. Results are:

Core: average 62°C, maximum 70.9°C
VRAM: average 66.1°C, maximum 74°C
Hotspot: average 70.4°C, maximum 81.6°C

I would say that is well within expectations and far from critical. So either they changed TIM since I bought it, Gainward is fine, or, as you said, it's just a certain batch gone bad. It definitely doesn't seem to affect all cards out in the wild.

 

[CmdrShepard]

I've considered Igor to be a reputable source but the number of times that he's the first to report on these issues and giving incomplete data (such as lack of country of origin, exact models, case layout, room temps, etc..) I take what he says with a grain of salt until Gamers Nexus or DerBauer have an analysis.

I'm not saying Igor is doing anything sketchy, it's just becoming a trend.

And as others have asserted, GPUs are NOT simple like they used to be and TH recommending people just tear them apart and repaste since it's trivial is wrong and a generalization. As pointed out there are thermal pads to consider during reassembly where down to the millimeter and quality of themselves could end up killing your card (like reusing them), microtears or misalignment you can't even see if you use the wrong ones.


At the very least if you are hell bent on doing it, have GOOD digital calipers, not HarborFreight HS (Harbor Freight for our European friends is , hit or miss, like buying your tools from McDonald's. Measure the pad somewhere where it hasn't been compressed. Take pictures, don't add pads to every component you see.

GPUs aren't maintenance free but I don't believe in "fixing"/rebuilding them until there is a reason to. Anyone with equipment expensive enough that they'd be mad if it failed should be running something like HWInfo64 at all times, knowing their system and noticing when there are abnormalities.

I'm not just saying that because my 4090 HASN'T exhibited issues but because there are things in life where the old "If it ain't broke don't fix it" isn't just a saying but actually a better idea.

You are all free to do as you like, I'm not a Steve or Roman (Edited; Don't know why I thought his name was Stefan), I'm just your average PC guy who's been in and out of the desktop scene for years who knows what works for them and what doesn't (or is destructive) from personal experience.


TLDR; Take Igors report and TH suggestion with a grain of salt.

That's all fine and dandy, but if your thermal paste is compromised that much for your GPU to develop a 100 C hotspot, then even if you repasted it and used the old thermal pads your risk of damaging the card would probably still be lower with the procedure than without it. Thermal pads cover VRMs and / or memory chips and those can actually be replaced if they fail. GPU on the other hand? Good luck with that.

 

[SyCoREAPER]

That's all fine and dandy, but if your thermal paste is compromised that much for your GPU to develop a 100 C hotspot, then even if you repasted it and used the old thermal pads your risk of damaging the card would probably still be lower with the procedure than without it. Thermal pads cover VRMs and / or memory chips and those can actually be replaced if they fail. GPU on the other hand? Good luck with that.

Go back and read what I said. I never said if you're having issues not to do it. I said doing this preventatively is a bad idea.

 

[umeng2002_2]

I dont understand the logic. Surely the amount of increased RMA's costs more than quality thermal paste right?

Business people aren't usually the brightest.

 

[cd_reader_eater]

Heh!
Makes sense. GPU die more quickly, customer buy new... = profit

And one cent save when selling millions is a lot of money too...

Yeah, this is bad, but really hard to find out without opening the GPU. And it is not something who is not really careful to do. You can break so many things, if you don´t know what you are doing,´. Not a hard thing to do, but cutting some wires in $3000+ GPU is not fun...

Then people will just RMA their GPUs after a few months under their warranty.

Doesn't make sense.

Also, why the heck would anyone buy a GPU again when theirs breaks after a few months?? This is not a game, people will get mad at the (almost) thousand dollar brick and never buy this garbage again.

 

[Silas Sanchez]

You have no business buying cards this expensive if you dont know how to repaste and repad. That might sound dogmatic but the alternative is to get screwed over and actually support the quality of things lowering and prices rising. Its not rocket science, the pads must be high quality designed for the task, as thin as possible, yet thick enough to bridge the gap between the two mating surfaces. A precision ruler is enough as most pads have generic sizes. The most important are the parallel bank mosfet switches, power mosfets are very sensitive to heat.

This is part of a more broad problem I have known about for a long time, my solution is to oversize the hardware. You dont buy a car with some wimpy European 2L engine, you buy one with a big 6L V8 petrol. Big and understressed and with lots of lovely pollution, that nasty bad stuff.

The real worry is things are getting harder to take apart and service, they dont want us servicing them. Greedy awful people in our society.

 

[beyondlogic]

Nvidia's AIB partners reportedly use dirt-cheap thermal paste that dries out in just a few months, causing many RTX 40-series graphics cards to overheat.

Nvidia GPU partners reportedly cheap out on thermal paste, causing 100C hotspot temperatures — cheap paste allegedly degrades in a few months : Read more

Already aware of this 4070 super is already hitting hot spots of 93 degrees. On my rig

 

[beyondlogic]

You can't just replace your Thermal paste. It's a lot more complicated. If you take it apart, you will need to replace all the thermal pads. Reuse is a bad idea. What thickness pads do you need ? Who knows? Better not get that wrong.
Kind of irresponsible to recommend folks tear apart their cards without more info on how complicated it can be..

It's complicates not impossible

I recommend digital calipers
Measure from uncompressed part of the pad.

Most cards use pads 0.5 1.0 1.5 2.0 and 3.0.
Yes it's costly to have them all but it's best to have them all on hand.

 

[Snarks Domain]

The real trick is to not use paste or pads at all. There are many good Phase Change Materials (PCM's) and Thermal Putties now widely available. PCM's can melt down to around 0.038mm or 0.0015". Using a Thermal Putty everywhere other than the core, instead of pads, allows more of the mounting pressure of the coolers spring screws/X-bracket to get that ideal core contact and the thinest layer of PCM possible.

TG-PP10 was the go-to Putty back in the day. When it stopped being sold I started hunting for alternatives. Here are some of the top putties I've tested

Upsiren UTP-8
Fehonda LTP81
Upsiren UX Pro Ultra
EVGA Putty
Penchem TH949-1
CX-H1300
Fehonda LTP65
Laird T-Putty 607
Halnsiye HY236/HY256

 

[Daniel15]

Why try fix it yourself? If it's a known issue then you should RMA the card and make the company fix it.

Depends on how strong the consumer protection is in your country. In Australia they'd have to repair, replace or refund, but consumer protection in the USA is significantly weaker so it may be harder to do.

 

[awake283]

Why try fix it yourself? If it's a known issue then you should RMA the card and make the company fix it.

Depends on how strong the consumer protection is in your country. In Australia they'd have to repair, replace or refund, but consumer protection in the USA is significantly weaker so it may be harder to do.

Cause people know how to do it themselves and they don't want to be without a GPU while they wait on an RMA.

 

[SyCoREAPER]

I tested my card all evening tonight. Bought the card 1 1/2 years ago, so if it had bad TIM like in the article it would have shown by now. The card in question is a Gainward Phoenix RTX 4070Ti. Results are:

Core: average 62°C, maximum 70.9°C
VRAM: average 66.1°C, maximum 74°C
Hotspot: average 70.4°C, maximum 81.6°C

I would say that is well within expectations and far from critical. So either they changed TIM since I bought it, Gainward is fine, or, as you said, it's just a certain batch gone bad. It definitely doesn't seem to affect all cards out in the wild.

Agreed.

Just played Ratchet and Clank maxed out (incl. Ray Tracing @4K (a very poorly optimized game);

GPU Temp: Max: 67.6 | Avg: 64.4
Memory Junc: Max; 80.0 | Avg: 75.4
Hotspot: Max: 77.3 | Avg: 72.4

I know the deltas are really close but my room is warm (it is summer) and as mentioned pushing the GPU to the limit.

Lower demanding games from previous monitoring, I only game on this PC and never leave it idle for more than a few mins (still don't trust the 12VHPWR connector unattended) during a 8h 52min session (must have really liked):

GPU Temp: Max: 68.4 | Avg: 61.8
Memory Junc: Max; 82.0 | Avg: 72.8
Hotspot: Max: 79.1 | Avg: 70.4


Lastly going back over a year, Feb 2023 (4h 49min session):

GPU Temp: Max: 69.9 | Avg: 65.4
Memory Junc: Max; 80.0 | Avg: 75.8
Hotspot: Max: 79.6 | Avg: 73.9


---
So in summary I agree with @KyaraM that it's batches or other unknowns. Monitor your temps and look for irregularities.
I've had my card since launch (+/- few weeks) and as far as I'm concerned, especially being a 4090, the temps are acceptable, well within spec and consistent in temps and deltas.

Edit: Forgot to mention I crossflashed the higher power limit BIOS and am also overclocked a tiny bit.

 

[cd_reader_eater]

You have no business buying cards this expensive if you dont know how to repaste and repad. That might sound dogmatic but the alternative is to get screwed over and actually support the quality of things lowering and prices rising. Its not rocket science, the pads must be high quality designed for the task, as thin as possible, yet thick enough to bridge the gap between the two mating surfaces. A precision ruler is enough as most pads have generic sizes. The most important are the parallel bank mosfet switches, power mosfets are very sensitive to heat.

This is part of a more broad problem I have known about for a long time, my solution is to oversize the hardware. You dont buy a car with some wimpy European 2L engine, you buy one with a big 6L V8 petrol. Big and understressed and with lots of lovely pollution, that nasty bad stuff.

The real worry is things are getting harder to take apart and service, they dont want us servicing them. Greedy awful people in our society.

Haven't read such gatekeeping elitist non-sense in quite a while.

 

[awake283]

You have no business buying cards this expensive if you dont know how to repaste and repad. That might sound dogmatic but the alternative is to get screwed over and actually support the quality of things lowering and prices rising. Its not rocket science, the pads must be high quality designed for the task, as thin as possible, yet thick enough to bridge the gap between the two mating surfaces. A precision ruler is enough as most pads have generic sizes. The most important are the parallel bank mosfet switches, power mosfets are very sensitive to heat.

This is part of a more broad problem I have known about for a long time, my solution is to oversize the hardware. You dont buy a car with some wimpy European 2L engine, you buy one with a big 6L V8 petrol. Big and understressed and with lots of lovely pollution, that nasty bad stuff.

The real worry is things are getting harder to take apart and service, they dont want us servicing them. Greedy awful people in our society.

I get your point but I totally disagree. The end user should not be required to know how to disassemble their card. What if a manufacturer sold you a car and said "if you dont know how to change a timing belt, don't buy this car"?