Thursday, October 27th 2022
NVIDIA Tells AICs to Collect RTX 4090 Cards with Burnt Power Connectors, Send Them to HQ
NVIDIA is responding to reports of the 12+4 pin ATX 12VHPWR power connector of its new GeForce RTX 4090 "Ada" graphics cards being unreliable, and posing a potential fire hazard. The company has reportedly instructed its add-in card (AIC) partners, companies that sell custom-design graphics cards; to collect all retail graphics cards with burnt power connectors, and send them over directly to NVIDIA HQ for investigation. Reports of the 12VHPWR connectors overheating due to improper terminal contact aren't new, but this is the first time a retail product implementing the connector is experiencing reliability issues.
It came to light when a Reddit user posted pictures of a melted 12VHPWR connector from an NVIDIA-supplied adapter that converts four 8-pin PCIe to one 600 W-capable 12VHPWR. There is also charring on the female connector on the card, but the user claims that the card is functional. Later this week, another Reddit user posted similar pictures of a burnt connector for their RTX 4090 card. NVIDIA director of global PR for GeForce, Bryan Del Rizzo, in a statement to The Verge, said that the company is in touch with the first owner who reported this problem, and is reaching out to the other, as part of their investigation.The GeForce RTX 4090 isn't just a thick graphics card, with air-cooled custom-design cards typically being 4 slots thick; but is also a "tall" card, with heights typically in the neighborhood of 150-160 mm. Add the 35 mm minimum clearance recommended for the 12VHPWR to not bend in order to function safely; and you have a total effective add-on card height requirement of 180-190 mm, which can be a very tight fit for most ATX mid-tower cases that offer a maximum CPU cooler height clearance of around 160-170 mm. A bending of the connector is almost a certainty.
Sources:
The Verge, Igor's Lab, VideoCardz
It came to light when a Reddit user posted pictures of a melted 12VHPWR connector from an NVIDIA-supplied adapter that converts four 8-pin PCIe to one 600 W-capable 12VHPWR. There is also charring on the female connector on the card, but the user claims that the card is functional. Later this week, another Reddit user posted similar pictures of a burnt connector for their RTX 4090 card. NVIDIA director of global PR for GeForce, Bryan Del Rizzo, in a statement to The Verge, said that the company is in touch with the first owner who reported this problem, and is reaching out to the other, as part of their investigation.The GeForce RTX 4090 isn't just a thick graphics card, with air-cooled custom-design cards typically being 4 slots thick; but is also a "tall" card, with heights typically in the neighborhood of 150-160 mm. Add the 35 mm minimum clearance recommended for the 12VHPWR to not bend in order to function safely; and you have a total effective add-on card height requirement of 180-190 mm, which can be a very tight fit for most ATX mid-tower cases that offer a maximum CPU cooler height clearance of around 160-170 mm. A bending of the connector is almost a certainty.
125 Comments on NVIDIA Tells AICs to Collect RTX 4090 Cards with Burnt Power Connectors, Send Them to HQ
For real:
I would rather believe that eVGA got to greedy, instead of charitable. Their BS doesn't add up.
What are we calling this?
Goes to show you how half-baked this new connector is, or they just never meant to have adapters with it. It should have been 8 positive contacts and 8 grounds, so adapters of such sort (4x 8-pin) could be safely used and even then with such big cards 90° is the only sensible option.
Yeah, their ARGB implementation sucks! It only glows in yellow and orange!
Interesting, over-engineered motherboards.
A surprisingly stellar sound card.
Decent peripherals.
Likely, non-nVidia graphics cards in the future.
www.jonpeddie.com/news/evga-wont-offer-nvidia-next-gen-series/
Maybe the way they operate just wasn't sustainable with the trend.
2) And completely non-competitive prices, bad availability
3) Just use a DAC, it's not 90s anymore
4) So does 1 million other brands
5) I will believe when I see it And yet only eVGA croaked so far, the previous OEM that croaked was Club3D, which didn't even sell nV GPUs. I hate these posts. They offer zero statistical insight, hardly anything useful and just pure hate. The reality is that graphics card brand hasn't mattered in decade or more and they are statistically indistinguishable. It's hilarious to see people so agitated about them.
Seriously?
Do you have even the faintest idea of how you would go about engineering a fix to an actual fire hazard? A hint: the statement "It's obvious how it fails" is very explicitly not a part of that process. You investigate everything as thoroughly as possible, to not only identify what failed - which might indeed be obvious on the surface - but to uncover as many speicfic details of how it failed and what that failure looked like as possible. You check if what seems obvious is actually correct. Then you check more samples to look for variance. Then you check even more to verify your findings. Then you test and see if you can trigger the same failure yourself. Then, having identified points and modes of failure, you start thinking of how to rectify the problem - updated production process, more protection somehow, or a new ground-up design. ... so they should have engineered the connector around an adapter that will only be relevant for a year, two at most? Yes, that sounds very sensible.
I really, really hope you're not a product designer or engineer, jeez. I'm not - nowhere close - but I guess I've absorbed enough of that kind of logic through osmosis or something to at least have some grasp of how these things work. Do they? I seem to remember quite a few bad things that were very much not covered up by Nvidia. Or am I somehow escaping their reality distortion field?
Realistically, I wanted peace of mind that I couldn't get from AliExpress/Amazon pre-made cables because I couldn't trust the crimp quality or the wire gauge used. On Corsair cables that come with their PSUs, I trust both - so I spent the time reconfiguring them to my needs. Having dismantled or reassembled multiple pins in almost 100 Mini-Fit Jr. plugs, I think it's fair to say that they are adequate but not more than that. The pins, the crimp mechanism, the receptacle, they're all fine for the ~4A per pair 12V pin they're rated for. I don't know if I'd trust them with more than 8A, and I sure as hell wouldn't be happy doing what Nvidia's done with their 600W adapter which, after Igor's lab teardown of the damn thing, seems to push up to 16.7A down two of the pairs.
I'm not sure about 8A per pin on the standard Mini Fit Jr connector. Nvidia are using over 8A on every pin of the even smaller HPWR pins. That's bad enough IMO but the fact that they've doubled-up the last two because 4 connectors doesn't integer-divide into 6 pairs neatly means that almost 17A is being asked of a pin that might handle 5, on a good day.
I think the saving grace for Nvidia is that of all the 4090s sold, the number that are regularly drawing the full 600W is small, and of that, the number drawing 600W for long periods is even smaller.
Do you, or anyone else remember the cable management trick for folding IDE cables? You would have it up with a hair dryer which allows you to bend them in half without issue ( very very rare cable breaks). I wonder of those trying to bend these cables are not using the same/similar technique and just cold bending.
Jay just found out the original 3090Ti adaptors are fine ( pin to pin construction ) (08:35)
Maybe Nvidia thought themselves confident ? then comes the cheapout 4090 adaptors ( the soldering mess )