Nvidia 40-Series power connectors, melting, and burning.

[StDragon]

Ohms Law strikes again! There's simply not enough meat on the connectors to supply the amount of current it's rated for. Absolute engineering FAIL!!!

[starsmine]

56 minutes ago, Senzelian said:

Which standard? ATX? As in the 24-pin, EPS and PCI-e?

yes. those exactly are rated at 30 insertions, so the hubub about this one being 30 is bunk. People just didnt know the the PCI-e 6 pin and 8 pin were also 30 so they were surprised.

[AndreiArgeanu]

[Spotty]

3 hours ago, Uttamattamakin said:

The ATX 3.0 power connector aka 40 series power

It's called 12VHPWR. Not ATX 3.0. Not 40 series power connector. Not Nvidia power connector. 

 

I don't mean to be all "well aktually" on you over the name but youtubers like Jayz2cents and Linus Tech Tips keep getting the name wrong and it's now a pet peeve of mine.

 

3 hours ago, Uttamattamakin said:

Any ATX 3.0 connector connected to a ATX 2.0 card is only half smart. 

 

3 hours ago, Uttamattamakin said:

In my educated opinion, as a physicist, there is a fundamental flaw inherent in connecting ATX 3.0 to ATX2.0. 

ATX 3.0 and ATX 2.0 is completely meaningless in this context. There is no "ATX 3.0 connector" or "ATX 2.0 card".

 

3 hours ago, Uttamattamakin said:

That communication is what mitigates the physical issue that come with pushing more energy into a smaller, more delicate connector.   

Without any sense pins present in the connector the graphics card sees it as being limited to 150W max. The 'communication' via sense pins unlocks higher power limits. The two sense pins which determine max power draw of the cable do not need a connection to the power supply, they can just be grounded in the connector. If the sense pins are controlled by the power supply then they can't change power limits while the power supply is in use, it requires the system to be shut down. The optional power good signal may detect a fault if the current exceeds 55A or the voltage drops below 11V but that relies on both the PCIe device and power supply to support the pwr_stable signal, which afaik none of them do (I may be wrong though but I haven't heard any of them implementing it). Though, even if both card and power supply supported it that may not detect a fault if it's a single pin in the connector that is overloaded so it theoretically might not trigger in the situations reported here. 

 

SENSE0 - ground/open to signal graphics card max power draw from cable
SENSE1 - ground/open to signal graphics card max power draw from cable
CARD_PWR_STABLE (optional) - provides power good signal from graphics card to power supply
CARD_CBL_PRES# (optional) - provides signal to power supply that cable is plugged in to a 12VHPWR device, for power supplies with multiple 12VHPWR cables can be used to limit power draw from 12VHPWR cables if multiple cables are in use. 

Edited October 25, 2022 by Spotty

[leadeater]

4 hours ago, Uttamattamakin said:

ATX 3.0 needs that bi directional communication to be safe. That communication is what mitigates the physical issue that come with pushing more energy into a smaller, more delicate connector.   

That data channel won't actually help here and it's actually in place anyway. The sense only happens at startup so if the cable is moved or jostled after POST the negotiated 600W will still be in place.

 

The converter cables have the ground sense pins which tells the GPU what power mode the 12 pin connector is in.

 

The connectors are just too small for 600W, at least that is my opinion anyway.

[Quackers101]

jay2cent and others have messed around with those "sense" pins.

from 600w ish, to 450/400w if missing then maybe to 350/300w or OFF or not enough power to boot, even if all power cables are correct with correct wattage supported.

[Kisai]

I'm just gonna say "Told You So" to whoever it was who thought this wouldn't happen.

Nvidia, Intel, PSU manufacturers, that one guy on this forum who thought I was joking.

 

Here's the thing. It's too late for the 40 series. Unless you explicitly buy a card that does not use this power connector, it's not a question of IF it will happen, it's a question of when. Notice how the pictures show the pins that melted are near the edge? What does that tell us? Think about how people handle cables.

 

Yes. They're melting because they're being pulled or pushed by the wire and not the physical connector. And how can we fix this? Well a bandaid would be be to make the entire connector 2" long (eg heat shrink the entire thing two inches from the connector) so there is nothing to "bend", but that doesn't solve the GPU-side or PSU-side of the problem, it just makes it somewhat less likely to happen by making it harder to bend close to the plug. Without redesigning GPU or PSU's, the only option here is really to replace the connector. At least with something with more surface area. The connectors we had before were fine, they just weren't "tidy"

 

And I'll repeat again for that one guy who thought I was joking. This is why people will avoid buying the 40-series. Sure, people are going to buy them, but at this point, there have been recalls for this exact kind of scenario. Even from NVIDIA. Everyone who buys a GPU and "melts" or "catches fire" (remember the RTX 3080Ti's catching fire?) should be contacting US Consumer Product Safety Commission.

 

[thrasher_565]

one reason i got a evga 3090... i was skeptical of that 12pin...

[StDragon]

55 minutes ago, Kisai said:

I'm just gonna say "Told You So" to whoever it was who thought this wouldn't happen.

Nvidia, Intel, PSU manufacturers, that one guy on this forum who thought I was joking.

I'm right there with you. and said as much on Sept 27th.

[Uttamattamakin]

3 hours ago, Spotty said:

It's called 12VHPWR. Not ATX 3.0. Not 40 series power connector. Not Nvidia power connector. 

 

I don't mean to be all "well aktually" on you over the name but youtubers like Jayz2cents and Linus Tech Tips keep getting the name wrong and it's now a pet peeve of mine.

Ok for the sake of normal people finding the information what does calling it 12VHPWR achieve?  They are the connectors found on Nvidia 40 series and ATX 3.0 power supplies.   That's what the part buying public needs to know. 

How about if i say "well akctually" current itself does not flow in wires like some kind of fluid.   It is really a current density in a wire, not even a flow of electrons through a wire, but really an interaction at the field theory level.  Classically or quantum field theoretically the energy is transmitted by the fields involved.  A point where the wires barely make contact is also a point where the energy of the E and M fields are concentrated and focused.  Hence heating and possible fire.*    

 

 

3 hours ago, Spotty said:

 

ATX 3.0 and ATX 2.0 is completely meaningless in this context. There is no "ATX 3.0 connector" or "ATX 2.0 card".

 

Without any sense pins present in the connector the graphics card sees it as being limited to 150W max. The 'communication' via sense pins unlocks higher power limits. The two sense pins which determine max power draw of the cable do not need a connection to the power supply, they can just be grounded in the connector. If the sense pins are controlled by the power supply then they can't change power limits while the power supply is in use, it requires the system to be shut down. The optional power good signal may detect a fault if the current exceeds 55A or the voltage drops below 11V but that relies on both the PCIe device and power supply to support the pwr_stable signal, which afaik none of them do (I may be wrong though but I haven't heard any of them implementing it). Though, even if both card and power supply supported it that may not detect a fault if it's a single pin in the connector that is overloaded so it theoretically might not trigger in the situations reported here. 

 

SENSE0 - ground/open to signal graphics card max power draw from cable
SENSE1 - ground/open to signal graphics card max power draw from cable
CARD_PWR_STABLE (optional) - provides power good signal from graphics card to power supply
CARD_CBL_PRES# (optional) - provides signal to power supply that cable is plugged in to a 12VHPWR device, for power supplies with multiple 12VHPWR cables can be used to limit power draw from 12VHPWR cables if multiple cables are in use. 

Mmm hmm and suppose those sense pins IDK catch fire would that not trigger that the CARD_PWR is not stable or some such?  I think sensing if things have broken down to that extent has to be some part of the specification somewhere. If not that's even more of a flaw!

 

 

*That's not even a joke or me being obtuse.  I was a SMALL part of a back and forth between this guy and much more well known content creators.  I mostly write a science blog that is sometimes reported on as news... among other things. 

 

I still think his explanation is BS chiefly because hes trying to go on about field theory but not go all the way with it. At the bottom it's all just particles and probabilities. 

 

Which leads back to the topic we are on.  The odds that these connectors are all going to fail within ... what.. a week of release to the public and have it not be a design flaw are very low.  No consumer product should be that dangerous and not also come with requiring a special license to work with it.  You know like we do with cars or something. 

[Uttamattamakin]

2 hours ago, Kisai said:

I'm just gonna say "Told You So" to whoever it was who thought this wouldn't happen.

Nvidia, Intel, PSU manufacturers, that one guy on this forum who thought I was joking.

 

Here's the thing. It's too late for the 40 series. Unless you explicitly buy a card that does not use this power connector, it's not a question of IF it will happen, it's a question of when. Notice how the pictures show the pins that melted are near the edge? What does that tell us? Think about how people handle cables.

.......

 

And I'll repeat again for that one guy who thought I was joking. This is why people will avoid buying the 40-series. Sure, people are going to buy them, but at this point, there have been recalls for this exact kind of scenario. Even from NVIDIA. Everyone who buys a GPU and "melts" or "catches fire" (remember the RTX 3080Ti's catching fire?) should be contacting US Consumer Product Safety Commission.

 

I too called out that this generation was going to be a cluster.  Running so much power through even bigger silicon was destined to cause power and cooling issues.  The cards come with a computer attached not the other way around the way it is now.  They use most of the power of the system, and have far fewer wires to get that power into them.  I mean at what point do they just have to give up and put something that looks like the motherboard power connector on a GPU?  

Either that or we need a microarchitectural revolution in GPU's.  We need an ARM of GPU's to do for them what ARM did for CPU's.  Something to at least ... apply some pressure to be more efficient. 

[Shzzit]

Searching pcie 8 pin melted, showed thousands of results and hundreds pictures.

 

 

By all your logic, we need to get rid of 8 pin too. 

[starsmine]

I really want more data before going in on this. It feels like its news because its new right now. Like when electric cars catch fire, as if gas cars don't, even when electric car fires are less common, its NEW so therefore its news. So I hesitate to make ANY sort of definitive claims one way or the other.

there is already 10s of thousands of 4090s in the wild, so I dont know if the odds are terribly high or low based off the few melts we have seen so far.

[johnt]

I'm surprised Nvidia didn't just angle the connector toward the front of the case so the bends didn't have to be as severe. Anybody want to join me in designing and selling 90-degree adapters for 12-pin?? 4090 READY!!

[sounds]

1 hour ago, Uttamattamakin said:

How about if i say "well akctually" current itself does not flow in wires like some kind of fluid.   It is really a current density in a wire, not even a flow of electrons through a wire, but really an interaction at the field theory level.  Classically or quantum field theoretically the energy is transmitted by the fields involved.  A point where the wires barely make contact is also a point where the energy of the E and M fields are concentrated and focused.  Hence heating and possible fire.*    

That's the thing about flowing fluids. It's really an instantaneous Navier Stokes solution for continuity, velocity, convection, pressure gradients, external forces and diffusion, mediated by the interactions of electromagnetic fields of all the atoms - electrons in orbitals, not unbound electrons moving in a metal. still current and fluid flow both are really an interaction at the field theory level - all the same field, too. Right?

 

It's the same whether it lights up a light bulb (electromagnetic force) or moves atoms in a water pipe (Coulomb forces, which are electromagnetic forces). At least common materials aren't significant electric generators the way they all exert Coulomb forces.

[Kisai]

58 minutes ago, Shzzit said:

Searching pcie 8 pin melted, showed thousands of results and hundreds pictures.

 

 

By all your logic, we need to get rid of 8 pin too. 

Find one that isn't a result of GPU mining and being cheap.

Almost every hit in google right now that isn't about the 4090 is from people trying do Ethereum mining. The only ones that aren't, appear to stil be EVGA PSU's.

This forum:

EVGA forum

https://forums.evga.com/EVGA-POWER-SUPPLY-MELTED-CABLE-BURNED-RUINED-GPU-m2589605.aspx

 

That one is a "cablemod" so that's not a stock cable.

 

"Generic PCIe cables bought off eBay", says enough.

 

 

A Seasonic staffer responded to this one:

https://www.overclockers.com/forums/threads/would-you-guess-gpu-or-psu-is-the-reason-for-pci-e-cables-melting.783429/

 

 

 

And then the image speaks for itself here

 

 

 

So pretty much we can make the association that:

a) cryptominers are idiots and/or cheap and their extender (molex/sata/splitter to pcie) cables are what's melting

b) EVGA blundered and shipped bad cables (but if you look between the details it seems that the ones I liked to, the user replaced the GPU but reused the PSU, so it's just as likely the torqued the wires, and maybe the EVGA cables are crap or from the wrong model.)

c) Users or the OEM used wrong/underrated cables

d) Users unplugged/replugged/sharply bent the cable, possibly by the wires

 

Of these, the only thing in common with the 12vhpr/4090 is D. 

 

What makes the story a bigger deal is that nvidia knew about the problem before releasing the FE cards. This there was ample time to "fix" it one way or another

Option 1: replace the GPU-side connector with something bigger, albeit proprietary, and ship a cable that has 4 PCIe 8-pins on it with no less than 5" of cable between the connector ends.

Option 2: permanently solder a cable with heavier gauge wire to those pins that doesn't melt.

Option 3: permanently attach the 12vhpr connector to the FE card on a rigid pigtail with heavier gauge wire to the GPU cooler so it can't be bent. This doesn't solve the PSU end, but it moves the problem off the card itself so it's an easier fix.

 

It's obviously too late to re-engineer the coolers to put four PCIe connectors on them. AIB's might still be able to.

[LAwLz]

Look, you can't just quote "the laws of physics" in this instance unless you actually do the math to show that it is too much power over too small of an area.

 

Yes, "more energy packed into a smaller container results in more thermal energy release" may be right, but that does not matter unless it reaches a certain point. The important question is, have we reached that point? Nobody in this thread that are "quoting the laws of physics" have actually done the math to prove that we have or haven't reached that point.

 

Quoting "the laws of physics" in this case is just as silly as saying "sorry dude, but it is impossible to lift a dumbbell that weighs 10 pounds. I can lift a 5 pound dumbbell but the laws of physics says that a 10 pound one will require twice as much energy to lift. It's literally impossible according to the laws of physics bro".

Yes, the laws of physics dictates that lifting a 10 pound weight requires more energy than lifting a 5 pound weight. But that does not mean lifting a 10 pound weight is a bad idea, nor that it is hard or impossible. It's not "it requires more energy" that is important. What's important is "where is the line between doable and not doable".

 

Quoting "the laws of physics" doesn't make you sound smart. It makes you sound dumb when you apply it incorrectly.

 

 

 

Anyway, my 2 cents is that we should probably wait and see how things play out. A lot of people seemed to have picked sides on this before anything bad had even happened, and those people are now very eager to prove everyone that they were right all along. 

I wish people would do a bit more research before swearing legions to one side of the debate. People brining up things like "it's only rated for 30 insertions so clearly it is a bad connector" is just making the anti-12VHPWR side look kind of dumb because the other power connectors are also just rated for 30 insertions.

 

Your first instinct to hearing the 30 insertions number shouldn't be "wow that's low, and therefore it's bad!". Your first thought should have been "that sounds low, but I don't know what the old and trusted connectors are rated for so I should probably look that up before making up my mind that this is bad".

 

As for these pictures about the melted connectors... It looks bad right now but I would like to get some more info on the story, and maybe wait for a few more occurrences before panicking. I have seen plenty of pictures of melted and burnt 6 and 8 pin power connectors in my days and I have never seen anyone rally to ban those or calling them a fire hazard. It is worrying that it seems to be (but not confirmed) quite a few instances of it happening in a short amount of time though. 

But I have also heard of people putting their phones in a microwave and then claim it just spontaneously combusted. So I am kind of skeptical against these types of claims in general. It is usually important to get the full story, because people often lie to seem more like a victim than they might actually be.

 

 

 

Also, lol at the people who are trying to push this as a conspiracy that Nvidia are trying to sell more adapters. Come the fuck on... I shouldn't even have to explain how silly that is. But as soon as something happens revolving Nvidia there is always a handful of people who will try and spin it as "Nvidia is evil!". 

The conspiracy theory doesn't even make sense to begin with because they are bundling the adapters with the cards. I don't even think Nvidia sells the adapters.

[Chiyawa]

Whoever green-lit this unsafe contraption in PCI-SIG should get 'fired'.

 

Anyway, long story short, pushing so much current into a wire is a disaster waiting to be happen. We should like increase the voltage to reduce current so the cable won't run that hot. I wonder which engineer failed their electrical engineering yet still get accepted by PCI-SIG just because...

[Ravendarat]

So Im not gonna pretend to be an expert on PC power supplies, but I have worked in Car Audio/ accessory installations for over 20 years and as soon as I saw that new connector I was concerned. Ive seen way to many instances of damage caused from under specing power wire ran to amplifiers in cars that have caused melted fuseholders as well as melted wire or other damage to vehicles through negligence. This connector seemed to scream similar outcomes to me. As I said I am no expert but I am definitely gonna watch this with interest

[LAwLz]

12 minutes ago, Kisai said:

-snip-

There are plenty of pictures of burnt/melted 6 pin and 8 pin power connectors on the Internet.

You can't just dismiss the examples you find because "it doesn't count if it's from EVGA, and it doesn't count if they mined on it, and it doesn't count if X, Y and Z".

 

What if the person who posted this picture had mined on their 4090?

 

 

Anyway, here are examples of burnt or melted power connectors that do not appear to be related to EVGA power supplies or GPU mining:

https://forums.tomshardware.com/threads/burned-pci-e-connectors-to-gpu.3426306/

https://forums.tomshardware.com/threads/motherboard-melted-my-power-supply-cable-is-there-a-solution.3517133/

https://www.overclock.net/threads/burnt-gpu-cable-on-psu-end.1513960/

 

 

 

I could post more but I think you get the point.

[Dogzilla07]

2 hours ago, Uttamattamakin said:

Mmm hmm and suppose those sense pins IDK catch fire would that not trigger that the CARD_PWR is not stable or some such?  I think sensing if things have broken down to that extent has to be some part of the specification somewhere. If not that's even more of a flaw!

Nope, no communication between the PSU and the GPU whatsover. No data transfer, nothing. Just 2 wires grounded and 4 states (4 binary combinations with the two wires).

 

CARD_PWR isn't used as of now, no cable/adapter/psu uses it, nor will use it for 4090. Only the Sense0 and Sense1 are used.

[Blademaster91]

17 minutes ago, Chiyawa said:

Whoever green-lit this unsafe contraption in PCI-SIG should get 'fired'.

 

Anyway, long story short, pushing so much current into a wire is a disaster waiting to be happen. We should like increase the voltage to reduce current so the cable won't run that hot. I wonder which engineer failed their electrical engineering yet still get accepted by PCI-SIG just because...

According to Gamers Nexus and Jaystwocents, PCI-SIG did send out a warning that the connector could overheat if they are bent too much, so it seems like Nvidia went with the power connector anyway, either Nvidia already had these cards manufactured or wanted a better looking connector on their FE cards.

I'm not surprised on this news though, I've said before I don't trust this connector and the wires are too small to safely handle 600W but people said I was wrong, and I'm not surprised people are still defending Nvidia on this connector even though its clearly flawed and Nvidia should just use the 8 pin connector.

[Kisai]

34 minutes ago, LAwLz said:

There are plenty of pictures of burnt/melted 6 pin and 8 pin power connectors on the Internet.

You can't just dismiss the examples you find because "it doesn't count if it's from EVGA, and it doesn't count if they mined on it, and it doesn't count if X, Y and Z".

 

What if the person who posted this picture had mined on their 4090?

I think you got the wrong message here. Every link I clicked on, was one of the 4 points I made with the overwhelmingly been people doing something stupid. Either mining on it with garbage cables, bending the cables, or reusing cables that have clearly been torqued.

 

 

34 minutes ago, LAwLz said:

 

Anyway, here are examples of burnt or melted power connectors that do not appear to be related to EVGA power supplies or GPU mining:

https://forums.tomshardware.com/threads/burned-pci-e-connectors-to-gpu.3426306/

Molex-to-PCIe, that was the expected outcome.

 

34 minutes ago, LAwLz said:

https://forums.tomshardware.com/threads/motherboard-melted-my-power-supply-cable-is-there-a-solution.3517133/

That's the motherboard EPS12v, not PCIe

 

34 minutes ago, LAwLz said:

 

 

https://www.overclock.net/threads/burnt-gpu-cable-on-psu-end.1513960/

That one, I believe is non-stock cables, they said they reused a GPU, so I hazard to guess they reused the PCIe cables from probably another system. So I can't judge either way.

 

34 minutes ago, LAwLz said:

Cable damage, or they torqued the cables when they were inside the chassis.

 

34 minutes ago, LAwLz said:

Also appears to have been a torqued cable.

 

34 minutes ago, LAwLz said:

 

I could post more but I think you get the point.

You seemed to have just cherry picked without reading the posts, but that's fine, the statement isn't "this never happens with PCIe connectors", it's "you have to have done something to the cable", and in every instance I've seen outside of the "mining" stories (which appear to be from using garbage) it's always reusing the cables.

 

Which if you've seen how people pull on cables (often by the wires, since the plastic itself has no leverage) that explains how. 

 

This 12VHPWR connector takes that problem and makes it immediately worse by having smaller gauge wires and pins. So while these stories usually result from "I was inside/upgraded my PC ..." immediately before the problem happens, the RTX4090 stores are new installs, PSU and GPU.

 

edit: Also, by having multiple cables (6/8 pin) it also makes it easier to unplug without torquing. The more pins there are, the more likliness you will torque them, even by accident when inserting it, since the connectors that aren't deep don't hit all the pins at the same time.

[Chiyawa]

1 minute ago, Blademaster91 said:

I'm not surprised on this news though, I've said before I don't trust this connector and the wires are too small to safely handle 600W but people said I was wrong, and I'm not surprised people are still defending Nvidia on this connector even though its clearly flawed and Nvidia should just use the 8 pin connector.

Yeah, I do feel you. Last time I recommend 750W PSU instead of 650W PSU for some one with RTX2060 and all the forum screw me a lot, then come the RTX3080. So yeah, sometimes many people just like didn't want to understand every single thing before writing.

 

4 minutes ago, Blademaster91 said:

According to Gamers Nexus and Jaystwocents, PCI-SIG did send out a warning that the connector could overheat if they are bent too much, so it seems like Nvidia went with the power connector anyway, either Nvidia already had these cards manufactured or wanted a better looking connector on their FE cards.

Hmm... Must have miss the memo there. But still, PCI-SIG should come up with a better connector rather than 12VHPWR one. If there is a risk, they should have abandoned it. For me, I would have use a 48V instead if I'm expecting that connector to carry 600W so the cable doesn't need to carry a lot of current that could cause the cable to overheat.

[Kisai]

16 minutes ago, Chiyawa said:

 

Hmm... Must have miss the memo there. But still, PCI-SIG should come up with a better connector rather than 12VHPWR one. If there is a risk, they should have abandoned it. For me, I would have use a 48V instead if I'm expecting that connector to carry 600W so the cable doesn't need to carry a lot of current that could cause the cable to overheat.

Won't happen, remember we're supposed to be moving to ATX12VO standard (which is actually the same problem if you think about it, removing a number of pins in favor of higher current over fewer pins.)

 

The best we can hope for is maybe a minor redesign of the PCIe slot + ATX12VO where 1200w can reasonably be pushed through the MB between the CPU and GPU, but even that to me seems like a suicidal direction. Perhaps the ATX12VO spec could instead be altered to not use this style of pin in the first place. eg, like a twist-lock or something.

 

Side note. a 600w inverter (eg 12V to 120V) actually use 9AWG wire for the 12V battery end. Now think about how thick that cable is compared to this 12VHPWR connector.