Another 16-pin RTX 4090 power adapter has melted, but from the PSU side this time

[Daniel Sims]

Why it matters: Since October, dozens of RTX 4090 owners have reported melting power adapter cables. Despite investigations from Nvidia and third parties, a definitive cause has yet to be determined. It was thought that pairing the GPU with an ATX 3.0 power supply was a safe solution… until now.

A new report of a burnt GeForce RTX 4090 power cable has emerged on Reddit, but this case is fundamentally different from prior incidents. If it's connected to the same flaw, it could cast doubt on earlier theories as to the problem's source.

User "Shiftyeyes67k" posted a picture of severe burn damage on a 16-pin 4090 12VHPWR power cable connector and the corresponding connection port on a 1000W BeQuet Dark Power 13 ATX 3.0 power supply. This case is potentially significant because despite looking similar to previous examples of scorched 4090 cables, it lacks two common factors between the others.

The few dozen cases that have emerged since October resulted in damage to the connectors plugged into the graphics card. Furthermore, they all affected customers who used the 4090's bundled 16-pin connection adapters to plug the GPU into an ATX 2.0 power supply's 12-pin socket.

Shiftyeyes67k bought an ATX 3.0 PSU with a 16-pin socket specifically to avoid this problem. Prior incidents led users to blame the adapter, and upgrading to an ATX 3.0 power supply is a small investment compared to the $1,600 4090, but this instance involved no adapter. Another owner who reported the problem last month had used a high-end third-party adapter, casting early doubt on Nvidia's included attachments as the source of the issue.

Nvidia, acknowledging about 50 cases worldwide in November, blamed improperly connected cables. Multiple affected users claimed they were confident their cables were fully connected, but others say the 12VHPWR connector easily loosens and can be less secure than it appears. Those wishing to take extra precautions can use a new feature in GPU-Z to check their GPU's power cable connection firmness.

Whatever the true cause is, it remains unresolved over half a year after Nvidia's last public statement regarding the issue. One user launched a class-action lawsuit against the company in November, and the reports have provided AMD with verbal ammunition in its rivalry with Nvidia. If more ATX 3.0 owners report damaged 4090 cables, it could point to a deeper problem with Team Green's flagship product.

Permalink to story.

https://www.techspot.com/news/99094-another-16-pin-rtx-4090-power-adapter-has.html

 

[SirDigby]

Is there any correlation or pattern with different PSU and GPU brands?

 

[Geralt]

Deserved. Only a mor0n could buy a 450W GPU. Maximum should be 350W. Nvidia is fully responsible for this dangerous mess.

 

[Achaios]

I agree with Geralt.

This is I think a non-issue as the ppl who can afford to buy a 4090 can afford to buy dozens of the most expensive insane 2000W PSUs too.

In addition 4090 owners are like 0.1% of the enthusiast user base. Not even "one percenters".

 

[hahahanoobs]

Now we wait to see what BeQuiet says about this.

 

[godrilla]

20 cablemods 16 pin cables were also burned via report past week FYI.

 

[DSirius]

The famous Youtuber ''specialist" Flexus will investigate, and decree that the user was too dumb and did not insert the power connector correctly. Thus, PSU manufacturer is quite innocent and Nvidia is Snow White innocent.

 

[takaozo]

This connector should come with thermal protection
If pin temp >= 70 Celsius then power off

 

[takaozo]

The famous Youtuber ''specialist" Flexus will investigate, and decree that the user was too dumb and did not insert the power connector correctly. Thus, PSU manufacturer is quite innocent and Nvidia is Snow White innocent.

The "tech Jesus" from GN will absolve again NV and PSU makers, "you are holding it wrong" rings any bell?

 

[godrilla]

The only way around this is to create a smart cable with thermal sensors on both ends that gives user feedback if not plugged in correctly. Color coded, digitally whatever.

 

[RandomWAN]

Deserved. Only a mor0n could buy a 450W GPU. Maximum should be 350W. Nvidia is fully responsible for this dangerous mess.

People plugging them in improperly is the cause. If this was an inherent design flaw, we'd see a much higher number of cases. No one has been able to replicate the issue outside of an improperly inserted plug.

Find the total 40 series cards sold and then the total number of confirmed cases of these plugs melting, then do the math. I bet you get a percentage way below 1%.

The "tech Jesus" from GN will absolve again NV and PSU makers, "you are holding it wrong" rings any bell?

Someone else needs reproduce the results then with a different cause then. No one else has show any alternative potential causes.

 

[Lew Zealand]

People plugging them in improperly is the cause. If this was an inherent design flaw, we'd see a much higher number of cases. No one has been able to replicate the issue outside of an improperly inserted plug.

How many times have 6 and 8-pin power connectors melted compared to how many thousands (millions?) of these are currently being used?

These 12-pin connectors seem to be more flawed than 6- and 8-pin, unless someone can show that this is happening as frequently per installed base with those older connectors.

 

[DSirius]

People plugging them in improperly is the cause. If this was an inherent design flaw, we'd see a much higher number of cases. No one has been able to replicate the issue outside of an improperly inserted plug.

Find the total 40 series cards sold and then the total number of confirmed cases of these plugs melting, then do the math. I bet you get a percentage way below 1%.



Someone else needs reproduce the results then with a different cause then. No one else has show any alternative potential causes.

Check here, no point in defending Nvidia. The 3rd video is creme a la creme.

 

[wiyosaya]

If this was an inherent design flaw, we'd see a much higher number of cases. No one has been able to replicate the issue outside of an improperly inserted plug.

For decades, there have been PC power plugs that were designed so that they could not be inserted improperly. Now we have PC power plugs that can be inserted improperly? IMO, this IS a design flaw.

In pre-ATX PS days, I had PC power connectors burn on me, and it was not possible to insert them incorrectly. Then why did they burn? Simple, they were not capable of handling the current flowing through them.

Designers these days, and back then, too, are interested in keeping costs low. I think its becoming blatantly obvious that these connectors, for whatever reason, are poorly designed. IMO, the designers need to re-evaluate the design and improve on it to make it impossible for this kind of thing to happen rather than try to pass the buck because they skimped on the design for whatever reason.

 

[RandomWAN]

For decades, there have been PC power plugs that were designed so that they could not be inserted improperly. Now we have PC power plugs that can be inserted improperly? IMO, this IS a design flaw.

In pre-ATX PS days, I had PC power connectors burn on me, and it was not possible to insert them incorrectly. Then why did they burn? Simple, they were not capable of handling the current flowing through them.

Designers these days, and back then, too, are interested in keeping costs low. I think its becoming blatantly obvious that these connectors, for whatever reason, are poorly designed. IMO, the designers need to re-evaluate the design and improve on it to make it impossible for this kind of thing to happen rather than try to pass the buck because they skimped on the design for whatever reason.

Most of them have always been able to be inserted improperly as in, not secured properly. I've run across many systems that had connectors not fully secured and this includes machines coming from major manufacturers of enterprise equipment. Now we have a glut of first time or low experience builders installing components and since the power draw is high enough, it's ending in failures.

At the end of the day, if 99%+ of people are not getting these failures and you cannot reproduce these failures outside of one failure mode, then that failure mode is the only option.

This is no different than when everyone was trying to blame failures of the operators on Toyota in regards to the unintentional acceleration. They couldn't reproduce these "failures" outside of engineering a failure by bypassing a number of the safeties.

If someone wants to blame this on the design itself, then they should be able to produce the failure where it isn't due to the plug not being properly inserted.

How many times have 6 and 8-pin power connectors melted compared to how many thousands (millions?) of these are currently being used?

These 12-pin connectors seem to be more flawed than 6- and 8-pin, unless someone can show that this is happening as frequently per installed base with those older connectors.

How many have had that much draw through a singular pin? The individual connectors are supporting enough draw to become destructive if you increase the resistance at the connection, so it become more important to ensure it's plugged in correctly. I've dealt with hands-on access to hundreds of machines and I've seen many that were pugged in incorrectly, but the draw was low enough that it didn't create a hazard.

The only thing I've been shown that needs to change is the location of the plug on the board, not the plug itself. There's hundreds of thousands of users at this point that are not having these failures.

 

[takaozo]

Someone else needs reproduce the results then with a different cause then. No one else has show any alternative potential causes.

Design issue as shown on Igor's Lab

The 12VHPWR connection finally gets an explicit recommendation for plug contacts in the standardization | igor´sLAB

It was long overdue, but is now finally official. In revision 2.01 of the ATX version 3.0 (Multi Rail Desktop Platform Power Supply) from February 2023, Intel has finally had the design of the…

www.igorslab.de

This is how the 12VHPWR is done correctly: Crimping instead of soldering and how to supply the GeForce RTX 4090 with power properly | igor´sLAB

First there were bent cables, then annoyed buyers of the at least 2000 euro expensive NVIDIA GeForce RTX 4090, no matter which model or manufacturer. The manufacturer of the 4-pin adapter from 12VHPWR…

www.igorslab.de

The horror has a face - NVIDIA’s hot 12VHPWR adapter for the GeForce RTX 4090 with a built-in breaking point | igor´sLAB

Those who are now beating up on the new 12VHPWR (although I don’t really like the part either) may generate nice traffic with it, but they simply haven’t recognized the actual problem with the…

www.igorslab.de

 

[Lew Zealand]

How many have had that much draw through a singular pin? The individual connectors are supporting enough draw to become destructive if you increase the resistance at the connection, so it become more important to ensure it's plugged in correctly. I've dealt with hands-on access to hundreds of machines and I've seen many that were pugged in incorrectly, but the draw was low enough that it didn't create a hazard.

Yes, exactly. They are not as fault-tolerant as the older connectors seem to be. That is a poor design choice when providing 400W or more through a single point of failure.

 

[Geralt]

Most of them have always been able to be inserted improperly as in, not secured properly. I've run across many systems that had connectors not fully secured and this includes machines coming from major manufacturers of enterprise equipment. Now we have a glut of first time or low experience builders installing components and since the power draw is high enough, it's ending in failures.

At the end of the day, if 99%+ of people are not getting these failures and you cannot reproduce these failures outside of one failure mode, then that failure mode is the only option.

This is no different than when everyone was trying to blame failures of the operators on Toyota in regards to the unintentional acceleration. They couldn't reproduce these "failures" outside of engineering a failure by bypassing a number of the safeties.

If someone wants to blame this on the design itself, then they should be able to produce the failure where it isn't due to the plug not being properly inserted.

For less than that car companies have recalled thousands of units. Why is Nvidia not recalling 4090 cards? The joke could burn your house if the computer happens to be unattended. It is serious.

 

[RandomWAN]

Design issue as shown on Igor's Lab

The 12VHPWR connection finally gets an explicit recommendation for plug contacts in the standardization | igor´sLAB

It was long overdue, but is now finally official. In revision 2.01 of the ATX version 3.0 (Multi Rail Desktop Platform Power Supply) from February 2023, Intel has finally had the design of the…

www.igorslab.de

This is how the 12VHPWR is done correctly: Crimping instead of soldering and how to supply the GeForce RTX 4090 with power properly | igor´sLAB

First there were bent cables, then annoyed buyers of the at least 2000 euro expensive NVIDIA GeForce RTX 4090, no matter which model or manufacturer. The manufacturer of the 4-pin adapter from 12VHPWR…

www.igorslab.de

The horror has a face - NVIDIA’s hot 12VHPWR adapter for the GeForce RTX 4090 with a built-in breaking point | igor´sLAB

Those who are now beating up on the new 12VHPWR (although I don’t really like the part either) may generate nice traffic with it, but they simply haven’t recognized the actual problem with the…

www.igorslab.de

Has he reproduced a failure? You can theorize all you want, but unless you're proving that with reproducible results, you're just guessing and proving nothing. The last time I saw anything from him, he was just speculating instead of actually doing real world testing that produced a repeatable failure mode.

Yes, exactly. They are not as fault-tolerant as the older connectors seem to be. That is a poor design choice when providing 400W or more through a single point of failure.

It's not a design failure, it just means people were able to get away with sloppy work previously. It's at a point where it takes 4 8-pin connectors, and the cards are already big enough with the larger cooling solutions they've resorted to. They would have to make the cards even larger than they already are to give enough board space to 4 8-pin connectors.

The only place where I would blame Nvidia is how they refuse to let board partners redesign the boards as they want and have to stick closely to Nvidia's board design. That's more of a management "failure" than an engineering failure.

 

[b3rdm4n]

People plugging them in improperly is the cause. If this was an inherent design flaw, we'd see a much higher number of cases. No one has been able to replicate the issue outside of an improperly inserted plug.

But isn't it so much easier and more satisfying to just **** on Nvidia and people who bought the 4090? those fools deserve it. /s

Yet to see any houses burn down or actual fire of any kind start, despite this massive unacceptable risk people latch onto. Could the connector be better? yes. Are all the burnt ones caused by user error? Also yes. I wonder how this will go when AMD take up the new connector.

 

[longjohn119]

Not Nvidia's fault and it's not a Nvidia 4090 connector unless it's on the card itself. Most importantly it's not even a Nvidia design, it's an Intel design for an Intel standard forced on the PCIe Group. 12VHPWR I an Intel design for Intel's ATX 3.0 standard. Nvidia had NOTHING to do with it other than being an early adopter of the INTEL 12VHPWR standard

 

[longjohn119]

Now we wait to see what BeQuiet says about this.

Yup because the cable itself was provided by BeQuiet not Nvidia who only include an adapter

In other words it's not Nvidia's fault when it happened on a non-Nvidia power supply using a cable that wasn't provided by Nvidia. The problem is with the Intel standard itself and thus needs to be solved by Intel

 

[DSirius]

Most of them have always been able to be inserted improperly as in, not secured properly. I've run across many systems that had connectors not fully secured and this includes machines coming from major manufacturers of enterprise equipment. Now we have a glut of first time or low experience builders installing components and since the power draw is high enough, it's ending in failures.

At the end of the day, if 99%+ of people are not getting these failures and you cannot reproduce these failures outside of one failure mode, then that failure mode is the only option.

This is no different than when everyone was trying to blame failures of the operators on Toyota in regards to the unintentional acceleration. They couldn't reproduce these "failures" outside of engineering a failure by bypassing a number of the safeties.

If someone wants to blame this on the design itself, then they should be able to produce the failure where it isn't due to the plug not being properly inserted.



How many have had that much draw through a singular pin? The individual connectors are supporting enough draw to become destructive if you increase the resistance at the connection, so it become more important to ensure it's plugged in correctly. I've dealt with hands-on access to hundreds of machines and I've seen many that were pugged in incorrectly, but the draw was low enough that it didn't create a hazard.

The only thing I've been shown that needs to change is the location of the plug on the board, not the plug itself. There's hundreds of thousands of users at this point that are not having these failures.

Lame try to justify a design flaw. You contradict your own affirmation and justifications.

How many have had that much draw through a singular pin?

Well, 8 pin power connectors do not have this issues because they were designed better than Nvidia 16 Pin power connector, especially TO NOT DRAW so much power (though they can), and were made especially more sturdier and power safe. This is a good and safe design.
Now Nvidia designed a flawed 16-pin 4090 12VHPWR connector.
Because they raise the power draw to 450 W but instead made a tiny 16 pin power connector.
This is the pinnacle of dumbness.
More, Nvidia made the hugest video card with it's 4090 FE than any previous videocard, and they put it a tiny 16 pin Power connector. It is not that they do not have enough space to have 3 or 4, 8 pin power connectors. They did it with 3090ti and had no issue at all.
Anybody with common sense can do the math and draws some intelligent conclusions.
I advise you to rely on facts and compare how was before and how is after, like with 3 and 4 - 8 pin power connectors there were NO issues, but with Nvidia 16 pin power connections, the systems are burning literally.
The rest is mental gymnastic and a lame damage control for Nvidia PR.

 

[DSirius]

Not Nvidia's fault and it's not a Nvidia 4090 connector unless it's on the card itself. Most importantly it's not even a Nvidia design, it's an Intel design for an Intel standard forced on the PCIe Group. 12VHPWR I an Intel design for Intel's ATX 3.0 standard. Nvidia had NOTHING to do with it other than being an early adopter of the INTEL 12VHPWR standard

I advise you to do a proper research. Nvidia designed, heavily sponsored and pushed 16pin power connector to PCI-SIG.

Let's talk about the 12VHPWR connector....

PCI-SIG is involved too but the real culprit is Nvidia.
After that, PCI-SIG and Intel revised the flawed design of Nvidia 16pin power connector, which confirm again that 1st Nvidia 16pin power connector design is flawed.

 

[takaozo]

Has he reproduced a failure? You can theorize all you want, but unless you're proving that with reproducible results, you're just guessing and proving nothing. The last time I saw anything from him, he was just speculating instead of actually doing real world testing that produced a repeatable failure mode.

Let me tell you what you might not know about me. I have a degree in Electronics and more than 15 years of design and implementation of electrical circuits for buildings office and residence. My former authorization was for low voltage circuits (under 1000V). I do not practice this anymore. Moved to an "office job" now.

What Tech Jesus has "prove" is that from 5mm contact area using only 50% of the surface will get the same result. IMHO this is forcing the result and not proper testing.

I have seen so many burnt out cables/outlets/distribution boxes to know that .
Let me tell you that the design of this new connector is not as good as it should be.
One pair of 16 AWG cables 7-24 cores will get you a max of 7 A @30 Celsius.

AWG Table

New connector has 6 pairs and this is like resistors in parallel.

2 pairs = 10.5A
3 pairs = 15.75A
4 pairs = 23.62A
5 pairs = 35.43A
6 pairs = 53.15A
Now converting to Watt, 53.15x12 = 637.85W
When you design an electrical circuit some spare must be added, usual calculation is 2/3 from max power = nominal power. This means that for 200W load you design 300W circuit.

Now you do the math and tell me is this OK or not? And this is only for wires.

Connector side now that 0.5 mm (<=) metal that the pins are made from

take the point where it's the thinnest and calculate power from AWG table related to resulting area. After that calculate the contact area and contact pressure.

Let me know what results you have.
For me this is the upper limit (600W) for this connector. Should not be used as nominal power. Now modern GPU's have power spikes like 200% or more. To be on the safe side 350-400W max power for this design. All GPU's with this power target have no issues with this and only the 450W 4090 has it.