Melting RTX 4090 cables could be much more common than previously reported

[Daniel Sims]

A hot potato: GeForce RTX 4090 owners have frequently reported burned and melted power cable connectors for over a year. Nvidia acknowledged a few dozen cases worldwide, blamed the problem on user error, and addressed the issue by quietly updating the cables. However, a California repair shop disputes the company's claims that it's rare.

Computer repair shop Northridge Fix claims that far more GeForce RTX 4090 graphics cards suffer from melted power cables than Nvidia admits. Although the connectors in newer shipments appear to have solved the problem, the original cause is still hotly debated.

In a recent YouTube video (masthead), a Northridge Fix repairman claimed that the shop receives between 20 and 25 RTX 4090s with burned power cables per week, or around 100 per month. The video doesn't mention how long the business has been fixing melted GPUs, but its numbers far exceed the 50 cases Nvidia admitted to a year ago.

Team Green blamed the problem on users improperly plugging the cables into the side of the graphics card. However, Northridge Fix also disputes that presumption due to the high number of occurrences. Other customers, including the plaintiff in last year's class-action lawsuit against Nvidia, claimed to have correctly installed their power cables, suggesting a deeper flaw in 12VHPWR adapters.

The issue stems from the original adapters Nvidia shipped with the 4090 so users could install 8-pin cables from older ATX 2.0 power supply units into its 16-pin socket. In September 2022, PCI-SIG warned that the adapters could reach unsafe temperatures in specific situations, but Nvidia said it fixed the problem before the flagship GPU's launch.

Customers haven't reported problems since Nvidia updated its 12V-2x6 adapters in recent months. Third-party tests confirm that the new headers maintain relatively low temperatures even when improperly installed.

Those using the old cables should seriously consider installing the 4090 with an ATX 3.0 PSU, which natively supports 16-pin connectors without requiring the troublesome adapter and costs less than potentially replacing the GPU. At least one person reported a 16-pin cable that burned while plugged into an ATX 3.0 unit, but the occurrence appears far rarer than the melted 2.0 adapters.

Looking forward, Asus is testing a more permanent solution – GPUs that replace external cables with a PCIe x16 power connector on the motherboard. The cableless system supports at least 600W, and RTX 4070 GPUs employing it should appear by early 2024.

Permalink to story.

https://www.techspot.com/news/100827-melting-rtx-4090-cables-could-much-more-common.html

 

[Prosercunus]

I am very much doubting that 100 cases a month claim. PC Enthusiasts are all very active on the main social media forums and there is no way that number adds up for a single shop.

 

[Theinsanegamer]

The connector was always a POS. They should have just upgraded the 8 pin that already worked fine.

I am very much doubting that 100 cases a month claim. PC Enthusiasts are all very active on the main social media forums and there is no way that number adds up for a single shop.

The vast majority of PC gamers are not enthusiasts and likely dont know such sites exist.

 

[hahahanoobs]

100 per week for a year is a lot for one issue.
Clearly reports from other sources will follow.....

 

[hahahanoobs]

The majority of PC gamers are not enthusiasts and likely dont know such sites exist.

That's how we found out the first time.

 

[Puiu]

I'm reserving judgement until I see at least one more reputable source that can collaborate with this report.

 

[Strawman]

100 a month? Yeah, right. The numbers don't add up at all. Most people would just RMA the card. Most people don't even know the guy. The amount of people that would have the connector melt, not go through RMA, and know this guy to send him their card, yeah, no. It ain't happening.

 

[toooooot]

I mean, it is the fastest most power hungry card up to date. Maybe they wanted to make a monster so much and beat AMD mercilessly that they just said, yep this is the risk, it is not safe but this is the risk and we are willing to take.

rtx 5090 I am sure will be safer.

 

[loki1944]

Whatasurprise...gasp

 

[toooooot]

I am very much doubting that 100 cases a month claim. PC Enthusiasts are all very active on the main social media forums and there is no way that number adds up for a single shop.

But enthusiasts are a small number of all users. Enthusiasts would surely report, Non enthusiasts, I am certain, would not.

 

[loki1944]

100 a month? Yeah, right. The numbers don't add up at all. Most people would just RMA the card. Most people don't even know the guy. The amount of people that would have the connector melt, not go through RMA, and know this guy to send him their card, yeah, no. It ain't happening.

not if it cannot be RMA'd; like bought off ebay

 

[Strawman]

not if it cannot be RMA'd; like bought off ebay

Well how many bought off ebay and had a melted connector and sent it to this guy? 100 a month is a huge amount of cards.

 

[WhiteLeaff]

These could be all Zotac or "Bad brand" models without/with short warranty. Or a wave of problems coming to light after being denied the RMA.

 

[yRaz]

Well how many bought off ebay and had a melted connector and sent it to this guy? 100 a month is a huge amount of cards.

We're pretty late into the 4090 life cycle, we might be starting to see a wave of "properly installed" cards starting to fail with this new connector.

The connector is trash and there are 2 revisions to it, both of them never fixed the original issue, which is increasing power delivery by 4times into a smaller connector.

 

[i like foxes]

These could be all Zotac or "Bad brand" models without/with short warranty. Or a wave of problems coming to light after being denied the RMA.

No , 6 brands (at least) and Founders edition are reported with such an issue . No exception for the big brands .

 

[Dimitriid]

I think the more important part here is public perception and not actual incident rates. You see most people will think about this from an intuitive perspective, not a technical one and this is true even from PC enthusiasts but more so because people wrongly assume only seasoned PC Enthusiasts ever buy the 4090 which isn't true all of the time.

So establishing that both non-tech people and even some tech minded people will look at this intuitively, it is very clear to see how the situation at least looks like: previous gen cards took 2, 3 and sometimes even 4 physically bigger cables so a single, physically smaller cable is obviously not going to be enough.

The thing about adopting a new standard is that you better get it right the first time around otherwise people are not going to listen to technical explanation about how the actual wire gauge is sufficient for the power requirements and how you improved the actual connector to ensure proper separation and this is just a minor hiccup attributable to user error or unforeseen issues (I.e. Physically pulling on the power cable slightly when manipulating the PC while opening it, closing it, cleaning it, installing additional components, etc.)

In other words, the gut reaction will always be the monkey brain going 'New berries made me sick, this are poisonous!' so there's always going to be a significant number of people blaming the new cable design even if it can be shown to be a significant percentage of human error.

And that's without even touching on the fact that it can be both things: the new design might be sufficient but more susceptible to human error during or after the install.

 

[Strawman]

We're pretty late into the 4090 life cycle, we might be starting to see a wave of "properly installed" cards starting to fail with this new connector.

The connector is trash and there are 2 revisions to it, both of them never fixed the original issue, which is increasing power delivery by 4times into a smaller connector.

We are even later into the 3090ti lifecycle, where are all the damages connectors there? Apparently the connector isn't trash

 

[yRaz]

We are even later into the 3090ti lifecycle, where are all the damages connectors there? Apparently the connector isn't trash

Are you suggesting user error is less common on the 3090ti?

It also doesn't help that not many 3090ti's were ever made so this problem may very well exist and we just don't know about it.

 

[Strawman]

Are you suggesting user error is less common on the 3090ti?

It also doesn't help that not many 3090ti's were ever made so this problem may very well exist and we just don't know about it.

I'm suggesting that besides a few 4090s that weren't installed properly there is no problem to be found. The 100 cards he is supposedly receiving monthly is a fairytale.

 

[yRaz]

I'm suggesting that besides a few 4090s that weren't installed properly there is no problem to be found. The 100 cards he is supposedly receiving monthly is a fairytale.

Well I'm inclined to believe it because from the very start I thought pumping more power through less connectors is just a bad idea. Went from 300 watts in 16 pins to 600watts in 12 connectors.

But this will be fairly easy to confirm or deny.

Simply dismissing it outright is foolish. The claims of melted connectors never went away even after 2 revisions. So if this guy has a repair shop and this is what he's getting we should atleast hear what he has to say.

If nVidia did create a faulty connector, made a standard around it and it cost consumers millions in damages while potentially increasing the risk of fire in people's homes, they deserve to be held accountable. On the otherside, if this guy is using sensationalism to generate money with social media, he deserves to be made a fool of. But NorthridgeFix didn't just make this claim the other day, they have been posting about getting hundreds of fried 4090s for nearly 9 months now.

But with all that said, nVidia has done enough dumb things around the 12 pin connector that I would actually like some investigation into this. one reason I heard that may answer the 3090ti/4090 question is that the 4090 has been prone to transient spikes that are so severe that they shut powersupplies off or. The power problems of the 4090 have been making headlines during its entire production.


anyone who pays attention to my posts knows I thought a 600watt 12 pin connector was a dumb idea(a dumb idea no one asked for). It could be AMD using it, it could be Honda for all I care. The pins are smaller, there are less of them and they increased the current per pin from 20 watts per pin to 50 watts per pin. And they wanted to make them smaller? AMD says they aren't going to use it and it looks like Intel isn't either. I also don't see any accessories using it, either.

 

[Sovietrebel]

I think the more important part here is public perception and not actual incident rates. You see most people will think about this from an intuitive perspective, not a technical one and this is true even from PC enthusiasts but more so because people wrongly assume only seasoned PC Enthusiasts ever buy the 4090 which isn't true all of the time.

So establishing that both non-tech people and even some tech minded people will look at this intuitively, it is very clear to see how the situation at least looks like: previous gen cards took 2, 3 and sometimes even 4 physically bigger cables so a single, physically smaller cable is obviously not going to be enough.

The thing about adopting a new standard is that you better get it right the first time around otherwise people are not going to listen to technical explanation about how the actual wire gauge is sufficient for the power requirements and how you improved the actual connector to ensure proper separation and this is just a minor hiccup attributable to user error or unforeseen issues (I.e. Physically pulling on the power cable slightly when manipulating the PC while opening it, closing it, cleaning it, installing additional components, etc.)

In other words, the gut reaction will always be the monkey brain going 'New berries made me sick, this are poisonous!' so there's always going to be a significant number of people blaming the new cable design even if it can be shown to be a significant percentage of human error.

And that's without even touching on the fact that it can be both things: the new design might be sufficient but more susceptible to human error during or after the install.

If there is a significant percentage of human error, then there is a very big red flag pointing to a very poor cable design. Poka Yoke is the way..

 

[RandomWAN]

There's two things NorthridgeFix needs to prove: 1) The claim of 100 cards a month (with as much recording as he does, he should be able to show off the number of cards or something close to the total). 2) He needs to prove an alternative failure mode.

Everyone likes to talk out the side of their mouth about the supposed physics of failure and says, 'This is the cause!' without presenting any actual evidence. The only recreated failure mode at this point is user error. If you plug the cable incorrectly, you can melt the connectors. Everyone posits theories of connector interface styles, construction, and materials used, yet none of these claims are ever backed up with hard evidence.

 

[wiyosaya]

Nvidia will say anything to shirk responsibility.

rtx 5090 I am sure will be safer.

That's confidence.

 

[yRaz]

There's two things NorthridgeFix needs to prove: 1) The claim of 100 cards a month (with as much recording as he does, he should be able to show off the number of cards or something close to the total). 2) He needs to prove an alternative failure mode.

Everyone likes to talk out the side of their mouth about the supposed physics of failure and says, 'This is the cause!' without presenting any actual evidence. The only recreated failure mode at this point is user error. If you plug the cable incorrectly, you can melt the connectors. Everyone posits theories of connector interface styles, construction, and materials used, yet none of these claims are ever backed up with hard evidence.

I've been saying for a long time that a huge function of a connector is to eliminate user error. If it fails in that function then it fails as a connector. That's the most basic argument I use. The other one I use is that it seems like a stupid idea to put more than double the power going through the connector while also making the contact area smaller.

I fail to see how either of those arguments are invalid.

Now someone else brought something up interesting earlier. Why wasn't this an issue on the 30 series? If nVidia changed the manufacturing process of the connector leading to a higher rate of user error then it is absolutely their fault.

 

[Theinsanegamer]

I've been saying for a long time that a huge function of a connector is to eliminate user error. If it fails in that function then it fails as a connector. That's the most basic argument I use. The other one I use is that it seems like a stupid idea to put more than double the power going through the connector while also making the contact area smaller.

I fail to see how either of those arguments are invalid.

Now someone else brought something up interesting earlier. Why wasn't this an issue on the 30 series? If nVidia changed the manufacturing process of the connector leading to a higher rate of user error then it is absolutely their fault.

It suggests they HAD to change something, there is no reason 3090/TI customers wouldnt be susceptible to user error but 4090 consumers would. These groups are by and large the same group of people. And the 3090ti was more power hungry.