Nvidia GPU partners reportedly cheap out on thermal paste, causing 100C hotspot temperatures — cheap paste allegedly degrades in a few months
Nvidia GPU partners reportedly cheap out on thermal paste, causing 100C hotspot temperatures — cheap paste allegedly degrades in a few months
Most (if not all) AIB partners allegedly use this dirt-cheap paste that looses performance in months.
I miss EVGA. :(
Are there any reputable vendors out there nowadays? I hope to keep my EVGA 3090 for as long as possible, but don't know where to turn when I eventually upgrade... Maybe just buy a card from another vendor and re-do the thermal paste myself?
So here's the thing, really: there's a lot of companies that make good hardware.
The problem is there's not a single remaining AIB that has above shit-tier support if something goes wrong. They're all fucking awful to deal with, slow, and just suck. See: the recent ASUS support kerfuffle, except it's not just ASUS so much as every vendor in those same spaces.
EVGA is missed because their warranty support team was fucking stellar in a universe of otherwise wet diapers.
People will recommend this brand, or that brand, but I haven't heard any consistently good stories about any of them like I did with EVGA.
My brother had an MSI AMD card and got denied an RMA because the HDMI port was supposedly burned out, which makes no sense to me. They sent a picture to us, and to me it looked like someone jammed a soldering iron in there or something. 🤷♂️
I think you’d be fine just repasting whatever card you go with. My spouse has an EVGA RTX 3080 FTW3, and I have an MSI RTX 3080 Gaming Z Trio. Both cards have been fantastic for us. If I stick with Nvidia, I might try a different brand but so far I’ve been very happy with MSI.
I have no experience with AMD cards, but I’ve heard that Sapphire is highly regarded, probably great like EVGA.
Intel I don’t trust yet, so I’ll wait before I consider their Arc cards.
Gainward. Always have been.
Enshittify my thermal paste to save 10 cents on a $500 graphics card why don't you.
Why not leave off a few resistors or use leaky caps while you're at it.
Not that it makes it ok, but I'd bet a large amount of money the single biggest reason for this happening now is how much Nvidia has been squeezing their board partners. It's why we lost EVGA.
The answer is money, but it’s not material cost that’s driving these crappy thermal interface pads, but labor expenses (and I’d guess consistency too). Pick-and-place is absurdly fast at putting components onto a PCB, and if they can put the pre-cut pads onto the board that’s huge for a manufacturer.
It’s the difference between slapping a post-it note, or the dot/line/X/cross/etc method with grease. No contest that TIM pads win for them, any fallouts get handled via warranty.
I assume they got a CEO from a car company.
At least it wasn't a Boeing executive.
You'd run a benchmark test on your computer and die.
From the article:
"Most (if not all) AIB partners allegedly use this dirt-cheap paste that looses performance in months."
Does anyone even spell check the damn opening headline anymore?
Redoing the thermals with new pads and Thermal Grizzly Kryonaut turned my 114C crash happy RX 6700 XT into an 84C stable speed runner. Worth the effort for a used gpu.
Ah FFS, is this why my 7800XT is hitting like 90 degrees trying to run Final Fantasy 14?
114C? Lucky you even booted unless it's just part of the compute cluster
Kryonaut is the shit. My Vega 56 has been living under a water block for more than 6 years now and thanks to Kryonaut, the temperatures are exactly the same compared to when I'd just applied it.
Same here on the same gpu.
This is something that simply, should not happen. This kind of mishap should have been weeded out during R&D.
I won't say any more on it right now, and I definitely won't excuse the behaviour. I do, however, want to give advice to anyone affected.
If you can RMA, then do it. If you just don't want to deal with all that, and you want it fixed, whether you're directly affected, or if you just have a hot/slow GPU, I strongly recommend redoing the thermal material. You don't have to go all out with fancy phase change material or anything (though, that is definitely an option if you want to spend the money on it), but repasting shouldn't be difficult.
My recommendation is to do a small amount of research and try to find something that's not too expensive that is hopefully non-conductive, so any screw ups don't end your card. If you can get thermal pads that are the right size, you might as well replace those at the same time.
Most of the time, getting the cooler off is simply a matter of taking off the backplate, unbolting the cooler from the GPU chip location (remove the tension bracket), then carefully pulling it apart, and disconnecting any fan cables/RGB as you go.
Clean existing thermal compound off with tissue/paper towel, etc, then clean and polish the surfaces (both the GPU and the cooler side) using alcohol, generally isopropyl, and either a microfiber cloth or something else lint free. In a pinch, q-tips work. Both sides should be a near mirror finish when you're done, though, depending on the cooler, it may not have been machined to a near-mirror, so just clean it until your cleaning cloth/qtip comes away mostly or completely clean.
Once cleaned, apply new compound, fix any thermal pads and reassemble (reverse of disassembly). Be very careful when reattaching the tension bracket to move in a criss-cross or "x" pattern, always go opposite of whatever one you just tightened, and tighten everything just a little as you criss-cross the plate to ensure equal pressure across the cooler. Everything else should be trivial in terms of order.
Once everything is tightened and secure, reinstall the card and test.
How is this even possible??
9 out of 10 dentists recommend that paste brand!!The stock paste on my 2080ti was a joke. It would cause my hotspot temp to be >100 and the fans would go crazy. Repasted and now it's great. Pita though.
Well I just re-tested my MSI 4070 Super Gaming X Slim. I got it back in January, and I reckon ambient temperatures were 3-4°c lower then.
Running the same tests, GPU temperature max was 4.9°c higher, hot spot was 8.1°c higher. Not much of an increase when normalised for ambient, but still an increase. Funnily enough the memory was exactly the same as before, and that uses pads, which does point to the paste getting less efficient, rather than dust (of which there is some).
Disappointing, but not all that dramatic. Don't think I'll need to repaste any time soon.
Hey, what are the exact dimensions of the part with the thermal paste? I'd be curious know just how much area is being covered.
As for the AD104, it has a die size of 294.5 mm^2, 35.8 billion transistors, 7680 CUDA cores, and 48MB of L2.
https://www.tomshardware.com/news/nvidia-reveals-secrets-of-ada-lovelace-gpus