Info 12VHPWR PSA on cable bending

[amenx]

AMD exec recent tweet...

 

[Stuka87]

AMD exec recent tweet...

I am going to miss quips like this when twitter totally burns down, which may happen anytime now.

 

[NTMBK]

I am going to miss quips like this when twitter totally burns down, which may happen anytime now.

Why, are they running it on 4090s?

 

[The red spirit]

I don't think bringing up AMD losing a lawsuit over attempting to use the ALU count as the core count is really equivocal here. I mean... Nvidia isn't being sued for false advertising. Frankly, the GamersNexus report gives Nvidia the best ammunition that it needs in a lawsuit: "Failures have been attributed to user error." (I'm sure Nvidia has been looking into this themselves as well.)

Although, it does bring up an interesting question... if we think the 12VHPWR connector is just too haphazard to use, who should be at fault for it? I do like the idea of having the sense pins connect after the power/ground pins have already made proper contact. However, if the main issue is with latching, it seems like that's probably the biggest area that needs to be addressed. Even if the sense pins are moved to connect last, it may not help if the connector is still not latched in properly. Although, that makes me wonder what happens if the sense wires are impacted during use. Does the card only check them on boot-up or does it continuously monitor them? It seems like it's a boot-up only process.

nVidia claimed that cards will work just fine, so anythign less than that is a failure. frankly, we even saw how seemingly fully inserted connector wasn't actually fully plugged in, there's no feedback from it. Pins in connector seem to be less than adequate as they clearly shave off and that happens way sooner than it should be. So perhaps PCI-SIG made bad spec for connector, but then again nVidia is a part of PCI-SIG and not all adapter vendors have this problem, so... Also 12VHPWR was meant to be smart as in communicating with power supply about capabilities and adjusting GPU power requirements on the fly. That doesn't apply to adapters, but on ATX 3.0 PSUs that didn't work at all and there are still big power spikes (TPU did a review). All I can say is that there was a poor engineering decisions made when creating it and some OEMs probably made a bit beefier adapters, meanwhile others built it exactly to PCI-SIGs spec. However, nVidia should have done enough QA to test their own cards, especially 2k USD cards, which they seemingly didn't do and their initial warranty void threats imo was unprofessional behaviour.

 

[gorobei]

View attachment 71425
I think they had plenty of space for 3x8 pin. But the main thing with the adaptors is they allow for either 3x or 4x 8 pin. So nvidia can make one single chonker card that accepts different amounts of power all using the same single 12+4 pin connector. Saves them money and looks a lot better than 4x 8 pins on a card with one being optional or something else weird.

you would like to think so based on the outer housing, but if you look at the pcb it is a different story. (the 3080 rog has some perspective in the focal length while the 3080 founders is nearly orthogonal, but i've tried to match up the pcie pins as close as possible). the cut and paste of the 3x 8pins is roughly the area they would have to take up. unless nv changed their basic tried/tested layout, they would have to make an even bigger pcb or abandon the through card fan hole. but they cant lose the hole and keep temps down. they made the 4090 through card fan even bigger so they lost even more pcb area.

 

[In2Photos]

GN posted a follow up video after Nvidia posted an official comment.

50 documented failures out of 125,000 RTX4090 cards sold or a failure of 0.04%
Nvidia will expedite the RMA process for any cards affected, even if user error was the reason for failure.

 

[The red spirit]

GN posted a follow up video after Nvidia posted an official comment.

50 documented failures out of 125,000 RTX4090 cards sold or a failure of 0.04%
Nvidia will expedite the RMA process for any cards affected, even if user error was the reason for failure.

Looks suspiciously low, all things considered. I can't believe it's only 50 bad cables, when Reddit alone has 26 cases. Not to mention that sold cards from nVidia don't mean sold by retailer too. So now it's anyone's guess what the actaul failure rate is.

 

[H T C]

50 documented failures out of 125,000 RTX4090 cards sold or a failure of 0.04%

That many? Isn't that number WAY TOO HIGH considering the short amount of time since the card was launched?

 

[Stuka87]

Looks suspiciously low, all things considered. I can't believe it's only 50 bad cables, when Reddit alone has 26 cases. Not to mention that sold cards from nVidia don't mean sold by retailer too. So now it's anyone's guess what the actaul failure rate is.

I wonder if thats 50 4090 FE failures. And does not include cards from other manufacturers?

 

[In2Photos]

Looks suspiciously low, all things considered. I can't believe it's only 50 bad cables, when Reddit alone has 26 cases. Not to mention that sold cards from nVidia don't mean sold by retailer too. So now it's anyone's guess what the actaul failure rate is.

That many? Isn't that number WAY TOO HIGH considering the short amount of time since the card was launched?

I guess neither of you watched the video. Sales numbers include board partners.

 

[H T C]

I guess neither of you watched the video. Sales numbers include board partners.

I watched it, but after the post.

Still, is that manufactured cards in stores and / or sold or just sold? If the latter, then the failure rate is A LOT HIGHER than the 0.4 mentioned in the video.

 

[Stuka87]

I guess neither of you watched the video. Sales numbers include board partners.

View attachment 71456

View attachment 71457

Was at work, so was just glancing at the post, not the video. I have watched it now. But we don't know if its cards shipped, or cards in customers hands.

I still stand by my thought that the connector is a poor design. Failure rates may be very low, but its not going to go away. The only way to fix it, is to fix the design so that it has a very positive connection, and a clip that will properly hold the connector in place. Where pulling on the wire wont pull the cable out.

 

[sze5003]

Looks suspiciously low, all things considered. I can't believe it's only 50 bad cables, when Reddit alone has 26 cases. Not to mention that sold cards from nVidia don't mean sold by retailer too. So now it's anyone's guess what the actaul failure rate is.

Doesn't sound like from watching the video that Nvidia's statement said anything about bad cables. It sounded more like they said the cables are not being fully plugged in basically trying to say it's user error.

Now of course the variance in manufacturing I'm sure has something to do with it too and I'm sure out of those 50 a bunch could be qualified as manufactured badly as compared to the FE adapters.

I think they know the design itself is not the best and they don't yet have a proper design in place which is why I imagine they are saying they will expedite even AIB RMA's.

 

[sze5003]

I wonder if thats 50 4090 FE failures. And does not include cards from other manufacturers?

Steve said 50 globally so knowing that FE's are harder to get and usually not available everywhere outside of the US, I would imagine it also includes AIB cards.

 

[The red spirit]

I guess neither of you watched the video. Sales numbers include board partners.

View attachment 71456

View attachment 71457

I watched whole video, I still stand by what I said. BTW retailer isn't a board partner, it's a shop where you buy card.

 

[The red spirit]

Doesn't sound like from watching the video that Nvidia's statement said anything about bad cables. It sounded more like they said the cables are not being fully plugged in basically trying to say it's user error.

Now of course the variance in manufacturing I'm sure has something to do with it too and I'm sure out of those 50 a bunch could be qualified as manufactured badly as compared to the FE adapters.

I think they know the design itself is not the best and they don't yet have a proper design in place which is why I imagine they are saying they will expedite even AIB RMA's.

I highly doubt that. Nothing really stops nVidia from pulling a Gigabyte here.

 

[In2Photos]

BTW retailer isn't a board partner, it's a shop where you buy card.

I never said it was.

 

[DooKey]

Lots of axe grinding going on IMO. This problem is not an epidemic. I'm willing to bet PCI-SIG will propose an engineering change that reduces the likelihood of user error. That's how things work.

 

[Heartbreaker]

I still stand by my thought that the connector is a poor design. Failure rates may be very low, but its not going to go away. The only way to fix it, is to fix the design so that it has a very positive connection, and a clip that will properly hold the connector in place. Where pulling on the wire wont pull the cable out.

It's an industry standard connector that you can order in different pin counts. Problem could go away with a good enough awareness campaign (put a big sticker on the cable bag with warnings, instructions). Seat the cable properly and you have no issue. Only being very loose and pushed to the side really causes a problem.

With all the torture testing out there, I have zero worries about the cable/connector, even running all the power through just 2 pins causes no significant heating. You just have to seat it properly.

 

[CP5670]

It's funny how hard it is to buy a 4090 given all these cable burning issues. I have a couple of the stock tracker apps on my phone and it barely stays in stock for 5 seconds before disappearing. By the time I add it to the cart, it's gone.

 

[BFG10K]

50 documented failures out of 125,000 RTX4090 cards sold or a failure of 0.04%

Emphasis mine.

Wow, exactly 50 failures? Not 51. Not 57. But exactly 50. I want to know what they define as a "documented failure".

But let's say it's 50. And lets say a car that launched a month ago had 50 cases of smoking melting because the owner didn't push the gas nozzle in all the way when refueling, because it wasn't obvious how to do this. Would that be acceptable?

 

[Tup3x]

I don't see them burning on their own without user error or faulty adapter. Considering how bulky the adapter is it's pretty obvious that things can go bad really quickly if you don't push it in all the way. Where they made a pretty big mistake is how they designed the adapter. It should have been angled. Also it probably would have been better if the connector would be slightly larger. Better latching mechanism wouldn't hurt either.

 

[lopri]

I don't know why they don't bundle a 90 degree adapter. There's gotta be a cost-efficient solution to this.

 

[Heartbreaker]

I don't know why they don't bundle a 90 degree adapter. There's gotta be a cost-efficient solution to this.

That's not really the problem. The problem is not seating the connector properly.

 

[dr1337]

The problem is not seating the connector properly.

How come nobody seated a 3090ti wrong? Or any of the older 12 pins? You'd think with the crazy number of cards they shipped across all the miners and scalpers that someone would have burned a 30 series card too, yet the only burnt 30xx cards I've found online were standard 8pin connectors.

Anyone else remember the 3090 launch and some EVGA cards were frying power delivery caps on the back side? That problem came and went just like this power connector issue. Makes me think, it seems like it wasn't a connector problem but something that was solved in a board revision. Or it is a connector problem and the issues was a small batch of bad female receptacles, and nothing to do with the dongles?