Nvidia RTX 4090 power connectors melting?

Well for one, this melting/burning that is occurring is almost 100% not due to thermal issues with running current through those connectors, so all these videos showing FLIR images and what not aren't going to give you any answers i.e. "it makes too much heat and is melting plastic". When you get melting like this it's due to arcing, when electrons jump through the air from one conductor to another (aka arcing) it's going to cause areas that can be two orders of magnitude hotter in temperature i.e. more than sufficient enough to melt the plastics these things are made o
That seem quite the incredible claim, arcing from one conducter to an other with air and plastic between them when there is a lot of conducting stuff around (specially with the adapter).

to arc over 1cm you need around 10,000 volt, those pin try to stay at 12 volt.

Specially in the context of the PSI warning showing just how much current imbalance could occur with those connector.

Adapter-Scheme-Web-4-980x617.png
https://www.igorslab.de/wp-content/uploads/2022/10/Adapter-Scheme-Web-4-980x617.png

At least with the adapter it would be strange for the current the use air in between pin instead of the connection among them.

Much higher current in lower resistance-better connected pin was observed and calculated at least from my understanding
 
Last edited:
Well for one, this melting/burning that is occurring is almost 100% not due to thermal issues with running current through those connectors, so all these videos showing FLIR images and what not aren't going to give you any answers i.e. "it makes too much heat and is melting plastic". When you get melting like this it's due to arcing, when electrons jump through the air from one conductor to another (aka arcing) it's going to cause areas that can be two orders of magnitude hotter in temperature i.e. more than sufficient enough to melt the plastics these things are made of. Now on an image like one here it could be due to the female side of the connector somewhere inside the socket, but you should also see damage inside the socket too which not many of these seem to be showing off.

Now the 64 dollar question is how would the arcing happen? Well if there is dissimilar metals in the conductors thermal variations may cause them to expand just enough to create arc areas, especially if the conductors inside have sharp corners, this could be temperature related due to high current. If there's any sort of pinch or pull of the conductor due to cable bending that could also create gaps that allow this. Or yeah this could be self inflected damage to gain attention, get an easy RMA, etc.

That said, while I'm sure some of these are in fact "self inflicted" I feel there are too many instances for them all to be self inflicted. But all these clicks for dollars videos like JayzClicksToo really aren't understanding what could be damaging these connectors by showing a video of them recording in IR the connector, you're not going to see it get so hot that it melts plastic, hell you won't even be lucky enough to catch an arc of current zapping because it'll be very fast and it's under the plastic, only if it gets so hot that it starts melting will you actually see a spike in temperature.
Here is a pic of melted video card connectors from Reddit. Sorry if posted before.
https://www.reddit.com/r/nvidia/comments/yo246v/4090_rtx_gigabyte_adapter_cable_burned_in_the/
 

Attachments

  • 1667772930757.png
    1667772930757.png
    659.8 KB · Views: 0
  • 1667772948989.png
    1667772948989.png
    675.1 KB · Views: 0
Interesting from the megathread, there is one where the user shows the mount: https://new.reddit.com/r/nvidia/comments/ygstby/early_stage_adapter_melt_for_msi_gaming_trio_x3/.
This image shows the connector is not as tight as it could be (shouldn't see the sense pins metal): https://preview.redd.it/e19gipmetsw...bp&s=fabdbbe61ab2fe9059d9c5d3a3f73cb97662db9a

1667778662953.png

Compared to my Suprim X, which basically same card:
1667778932264.png

And another one where connector not completely plugged in (shoudn't see the individual plastic pins):
https://new.reddit.com/r/nvidia/comments/ygppat/another_16pin_adapter_melting_around_8hrs_total/
1667779203405.png

And another one (guess Instructions really need to point this out):
https://new.reddit.com/r/nvidia/comments/yfpjjl/a_friend_messaged_me_to_check_on_my_4090s/
1667779525229.png
 
Last edited:
to arc over 1cm you need around 10,000 volt, those pin try to stay at 12 volt.
It's a tad more complicated than that, I believe that value often gets thrown around between two conductors with no (or very little) current flow then electrons can hop the gap. When you're pushing lots of current that value no longer applies as electrons are trying to get away from each other and end up giving an extra "push" to other electrons requiring less voltage to hop gaps, it's the basics of how arc welding works (granted arc welders put in considerably more current than graphics cards), and it's less about the electric potential, and more about the electric field that is created, when conductors get "pointy" (needle heads, sharp corners, etc) the electric field from the charge on the conductors grows considerably technically it becomes infinite at a perfect corner but perfection doesn't exist so it just gets really high, again with arc welding it helps to have a pointy tip to get your zappy zap. Also typically you have a "pinch" with the connector as it's put in due to a taper or what not meaning the conductors do eventually touch with the gap between them getting smaller and smaller so much like a Jacob's ladder works where the arc starts where the conductors are closest and then can continue to get wider apart (granted in this case you have the opposite, low current but high voltage).

Granted it's been more than a couple decades since I've done JD Jackson electrodynamics but I'm fairly sure I'm getting the basic gist of things right, I know I'm a lot older and the memory doesn't work quite as well but hey I try :)
 
Has anyone tried an adapter from a 3090? They don't have the sense pins but that shouldn't be an issue.
 
It's a tad more complicated than that, I believe that value often gets thrown around between two conductors with no (or very little) current flow then electrons can hop the gap. When you're pushing lots of current that value no longer applies as electrons are trying to get away from each other and end up giving an extra "push" to other electrons requiring less voltage to hop gaps, it's the basics of how arc welding works (granted arc welders put in considerably more current than graphics cards), and it's less about the electric potential, and more about the electric field that is created, when conductors get "pointy" (needle heads, sharp corners, etc) the electric field from the charge on the conductors grows considerably technically it becomes infinite at a perfect corner but perfection doesn't exist so it just gets really high, again with arc welding it helps to have a pointy tip to get your zappy zap. Also typically you have a "pinch" with the connector as it's put in due to a taper or what not meaning the conductors do eventually touch with the gap between them getting smaller and smaller so much like a Jacob's ladder works where the arc starts where the conductors are closest and then can continue to get wider apart (granted in this case you have the opposite, low current but high voltage).

Granted it's been more than a couple decades since I've done JD Jackson electrodynamics but I'm fairly sure I'm getting the basic gist of things right, I know I'm a lot older and the memory doesn't work quite as well but hey I try :)
You seem to be right all around (and I did try to use the much smaller value of pointy active cable instead of the much larger usual value for that reason), has tension use in arc welding does not seem particular high.

Color me skeptical still. My 4090 is still perfectly fine after more than two weeks.
Skeptical of what, I feel it would be strange with all we seen and testing to be skeptic that the issue occur, you mean that the issue in a relevant frequency for perfectly connected, perfectly good cable ? Because a sample size of 1 over 2 weeks, is quite small.
 
It's a tad more complicated than that, I believe that value often gets thrown around between two conductors with no (or very little) current flow then electrons can hop the gap. When you're pushing lots of current that value no longer applies as electrons are trying to get away from each other and end up giving an extra "push" to other electrons requiring less voltage to hop gaps, it's the basics of how arc welding works (granted arc welders put in considerably more current than graphics cards), and it's less about the electric potential, and more about the electric field that is created, when conductors get "pointy" (needle heads, sharp corners, etc) the electric field from the charge on the conductors grows considerably technically it becomes infinite at a perfect corner but perfection doesn't exist so it just gets really high, again with arc welding it helps to have a pointy tip to get your zappy zap. Also typically you have a "pinch" with the connector as it's put in due to a taper or what not meaning the conductors do eventually touch with the gap between them getting smaller and smaller so much like a Jacob's ladder works where the arc starts where the conductors are closest and then can continue to get wider apart (granted in this case you have the opposite, low current but high voltage).

Granted it's been more than a couple decades since I've done JD Jackson electrodynamics but I'm fairly sure I'm getting the basic gist of things right, I know I'm a lot older and the memory doesn't work quite as well but hey I try :)
No.

Voltage is what causes arc's across a gap. ~30,000 volts per CM of gap. High current 'welding' at 12v requires contact.
Nothing about these melted conductors I have seen suggests that the metal contacts are 'melting' or 'welding together'. The plastic melted from heat of around 100C. Arc welding generates temperatures of around 3600C (or more), and that heat affecting electron flow is probably what you are remembering, but 100C =! 3600C.
You can arc weld with 24 volts and 1000 Amps, PSU's are nowhere near that.
 
When you're pushing lots of current that value no longer applies as electrons are trying to get away from each other and end up giving an extra "push" to other electrons requiring less voltage to hop gaps.
I haven't heard of this effect. Do you have a reference?
it's the basics of how arc welding works (granted arc welders put in considerably more current than graphics cards),
In my (very limited) experience, with arc welders you typically have to do an initial strike first to form the arc, after that you can draw the stick away and the arc maintains because the arc is much more conductive then regular air. It would make sense since arc welders usually use low voltage so the initial arc cannot form until you close the distance.

However, regardless of the 2 questions above, I still didn't fully understand your initial arcing post. You are suggesting what is arcing to what? Not +12V to GND right? I agree with LukeTbk, that would require substantial voltage.
 
I really hope anyone that has to leave their PC on all day isn't leaving it alone for hours at a time. Can yall imagine if people were mining with these cards. How many fires would have started by now?
 
I'm at 2 weeks since I got a 4090 and 1 week of nvidia 4 plug adapter and last week of cable mods evga 4 plug PSU cable. I've inspected the plug prior to posting this and it shows no signs of melting. PC is on 24/7. I've put 38 hours into spiderman remastered PC, and probably another 50 hours in Destiny 2. I think the thing I see common for all these melted pins is the fact it uses the split pin connector. This allows any side loading of the cable to increase the pin connector mating distance. Cable mod uses a single split female connector which when plugging into the slot feels much more solid in it's contact.
 
Yeah was worried at first because defective cables definitely sounds like a thing, but using both the included Nvidia adapter and a 3rd party Amazon one for past few weeks without any issues think most issues is user error (ie. loose connection for most then bending for the rest. I think defective adapters, ports (ie. excess plastic preventing proper connection, bad solder, wires etc.) would be the extreme case, but not completely dismissing until some official report comes out.

For now I think the combination of the 16 pin connector is default connector for most 4090s and it is does take some force to ensure proper connection makes the problem more common.
 
After unplugging it to take pics, I went to plug it back in, but without any significant feedback, I wasn't sure it went in completely. I'm glad I looked further as I noticed the backing piece cable mod provide can slip out from the cable giving a false sense that the connector moved down into place. I re-seated the backing guide piece and carefully replugged it, but it's something to note that on the aftermarket cables like cable mod which uses a wire guide that the wire guide backing piece is not firmly locked into place. On close inspection I was about to put myself into a partially plugged in scenario. In the picture you can see the additional piece wrapping around the plug. It just slides into place and when plugging it can slide down and off the plug if you are just gripping the outer part wrapped around the plug.

20221107_175228.jpg
 
Last edited:
I still don't understand why we couldn't just get four 8-pins. These boards are massive. God forbid, I know, it would cost them another 2 bucks to build the card with those extra connectors.
 
I still don't understand why we couldn't just get four 8-pins. These boards are massive. God forbid, I know, it would cost them another 2 bucks to build the card with those extra connectors.
How else are they going to get people to buy the new spec PSUs? Since MSI/Asus/etc. now also sell them, it only works to their benefit.
 
I haven't heard of this effect. Do you have a reference?
A quick google picked this up which gives an idea of what I'm talking about, granted it's a static case with an intro level of physics but it illustrates the whole "it's the voltage" argument and how that can be misinterpreted.

https://phys.libretexts.org/Bookshelves/University_Physics/Book:_Introductory_Physics_-_Building_Models_to_Describe_Our_World_(Martin_Neary_Rinaldo_and_Woodman)/18:_Electric_potential/18.04:_Electric_field_and_potential_at_the_surface_of_a_conductor
You can ignore most of this just scroll down to to figure 18.4.2 and read the paragraph above and below with conductors if you have sharp corners or "pointy bits" you get a larger charge distribution at those locations (see the picture of the knife) which leads to a much higher electric fields due to those charges at those locations, meaning "the voltage" between two points is going to be different based on the geometry of the conductors. When large amounts of current is flowing through you're naturally going to have more bunching up of charge hence the pointy bits with higher field strengths than the rest of the conductor. And yes I know when you strike an arc with a welder you "touch" to the surface but the point is you don't need to keep touching as the air is ionized from the heat, just used that as an example of not needing 3e4 volts per cm, but you're also not getting a sustained arc either if anything it's probably extinguishing itself.

But who knows, maybe there is some issue with resistance spiking causing high temps to melt the plastic, I see a that diagram of an improperly seated pin where you have a point of contact then a gap forms that slowly gets bigger with little pointy bits making the contact, I see arcing potential regardless of "the voltage".
 
A quick google picked this up which gives an idea of what I'm talking about, granted it's a static case with an intro level of physics but it illustrates the whole "it's the voltage" argument and how that can be misinterpreted.

https://phys.libretexts.org/Bookshelves/University_Physics/Book:_Introductory_Physics_-_Building_Models_to_Describe_Our_World_(Martin_Neary_Rinaldo_and_Woodman)/18:_Electric_potential/18.04:_Electric_field_and_potential_at_the_surface_of_a_conductor
You can ignore most of this just scroll down to to figure 18.4.2 and read the paragraph above and below with conductors if you have sharp corners or "pointy bits" you get a larger charge distribution at those locations (see the picture of the knife) which leads to a much higher electric fields due to those charges at those locations, meaning "the voltage" between two points is going to be different based on the geometry of the conductors. When large amounts of current is flowing through you're naturally going to have more bunching up of charge hence the pointy bits with higher field strengths than the rest of the conductor. And yes I know when you strike an arc with a welder you "touch" to the surface but the point is you don't need to keep touching as the air is ionized from the heat, just used that as an example of not needing 3e4 volts per cm, but you're also not getting a sustained arc either if anything it's probably extinguishing itself.

But who knows, maybe there is some issue with resistance spiking causing high temps to melt the plastic, I see a that diagram of an improperly seated pin where you have a point of contact then a gap forms that slowly gets bigger with little pointy bits making the contact, I see arcing potential regardless of "the voltage".
It’s almost like they’re little lightning rods :)
 
When large amounts of current is flowing through you're naturally going to have more bunching up of charge hence the pointy bits with higher field strengths than the rest of the conductor.
I'm familiar with the idea that charge is not distributed equally through a conductor and so pointy bits bleed charge easier. But I haven't heard of the effect where the amount of current going through the conductor will have an effect on this. Thats what I was interested in.

And yes I know when you strike an arc with a welder you "touch" to the surface but the point is you don't need to keep touching as the air is ionized from the heat, just used that as an example of not needing 3e4 volts per cm, but you're also not getting a sustained arc either if anything it's probably extinguishing itself.
Yes, but in the context of this connector issue the question is how would that arc form in the first place. In arc welding the arc is formed by contact. In an insulated plastic molded connector it's not clear to me why an arc would initiate. Not saying arcing isn't occuring, just trying to understand why you think this might be part of the issue and how it would occur.

edit: I'm also familiar with inductive kick causing very large voltage spikes. However I doubt there is enough inductance/current in a GPU system to cause the votlage to rise high to cause an arc. If there were I'm still not sure how this connector could cause such an issue (sudden disconnect during use?). So this seems unlikely to me as well.
 
Last edited:
Yes, but in the context of this connector issue the question is how would that arc form in the first place. In arc welding the arc is formed by contact. In an insulated plastic molded connector it's not clear to me why an arc would initiate. Not saying arcing isn't occuring, just trying to understand why you think this might be part of the issue and how it would occur.
756676_1667164958885.png


This image posted by Kyle on "page 5" (post #178) is what got my grey matter juices flowing. It's an exaggeration of how big the separation is, as in it could be in very close contact to get an arc going especially it if started near the pinched off part. The plastic sleeve surrounds both conductors so while the arcing is not directly melting the plastic it could heat up the conductors enough to melt it. Meaning it's pushing it past that resisitve threshold
 
You aren't going to get an arc off 12V. Needs 400V to jump a gap 0.03mm

Here's a nifty calculator:

https://www.cirris.com/learning-center/calculators/50-high-voltage-arc-gap-calculator

The image posted above showing a crooked connection is more likely to cause your atypical resistance problem. The GPU is going to pull x wattage from that wire but due to poor contact, the overall resistance at that junction is much higher. It will pull more current from less contact points, so in the case of the burnt connector, at the tip of the plug which matches with the melting occurring. I think due to the cheaper split female connector, any lateral load causes the split to grow wider allowing the pin inside to go in crooked in the picture and make the hot spots.
 
Last edited:
Balls, you can get sparks off 6V.

Sparks are not the same thing as arcing. The sparks are caused by a large amount of current creating heat as the surfaces come into contact. In contrast, static electricity for example is very high voltage which is why it can arc over a fair distance but does not create sparks because no significant current is being transferred.
 
756676_1667164958885.png


This image posted by Kyle on "page 5" (post #178) is what got my grey matter juices flowing. It's an exaggeration of how big the separation is, as in it could be in very close contact to get an arc going especially it if started near the pinched off part. The plastic sleeve surrounds both conductors so while the arcing is not directly melting the plastic it could heat up the conductors enough to melt it. Meaning it's pushing it past that resisitve threshold
Thanks for posting the picture. I wondered what made you think about sparks. The pointy surfaces are not going to accumulate significant charge in this scenario, but good thinking here.
 
I think sfsuphysics are on same page for looking root causes, which is probably multiple (user error, product defects (videocard connector, cable, psu, etc.) design, etc.). The root problem we're trying to figure out is what causes temps of 80c or above to melt plastic because 12 to 25 amps through short 16 gauge wire shouldn't cause that high of temps. Maybe amps above 25 over a sustained period (hours).

The prevailing cause (I'm assuming) for melted pci-e power connectors prior to 4090 is loose connections: https://electronics.stackexchange.c...ns-cause-overheating-arcing-and-start-a-fires. Loose connections cause a new series resistance R that over a very small space can cause enough wattage (current squared x R) through new resistance to cause localized temps of over 80c. If loose connections is cause for 4090 melting then causes could be user error, above diagram with side load, or defects in connector (bad plastic mould, bent metal pins, etc.).
 
Sparks are not the same thing as arcing.

So what, who gives a shit. Melting is melting.
Well, neither sparks nor arcing is what is occurring.

~30 Amps flowing thru only a few conductors, when it is intended to flow thru all 6, means the wires natural resistance = more heat than intended.

[Skip this math if you don't care]
W= IxV (W = Watts, I= Current, V= Voltage, R=Resistance), and V=IxR where this V = the voltage dropped across the natural resistance in the wire: Example: You got 12.00v available at the PSU output. You have 18 inches of wiring to the connector on the video card. 16AWG wire has a resistance of ~4 ohms per 1000 feet, or 0.0015 ohms in 18 inches of wire. Lets turn that into voltage drop with V=IxR, and estimate on the high side at 600W at 12V = 50A (in reality it is less because the pcie slot can do I think 150W of the total load). V=IxR so V=50x0.0015= 0.075 Volts dropped (lost, absorbed) by the 18 inches of wire. Convert that back to watts at 50Amps: 50 x 0.075 = 3.75 Watts. So if the GPU pulls 600W thru the 12vhpwr connector, the wires are absorbing 3.75W. And this is a complete worst case example, calculated for a single 16Awg wire supplying that 50 Amps. Now lets look at the 12vhpwr cable. 12 total wires to carry the load. 6 in one direction and 6 back. So the load, in a cable working as intended, is only carrying 1/6th of the current, or 8.33Amps. Only ~0.0125 V is dropped per wire, times 8.33A per wire = 0.1 W lost, per wire. Pretty significant decrease. So you can visualize what will happen if the scenario goes in the other direction.[/Math]

So where does it all go sideways?
Poor contact in that image above shows part of the suspected problem. The mating surface is supposed to be larger. In the math above I showed that the resistance drops when you add more conductors, in these badly made adapters, the reverse is happening. The resistance goes up because the size of the 'pipe' carrying the current gets reduced. The current requires fully open pipe the entire distance. That pin making poor contact is like a partially kinked garden hose, but one that is still moving the same amount of water. At the kink the pressure goes way up. Translating that back to electricity, that is the point of very high resistance. So R goes from very close to 0, to say 0.5 ohms (I don't know what it is actually going to, just guessing to make the example). 0.5/.0015(what the resistance should be) is an increase of 300x. 300x would also be the multiplier for the amount of heat lost at that point in the electrical path. So, going from .1W lost to 30W, you can see how quickly the power that is dissipated fighting the resistance goes up. The resistance at the poor contact points is likely somewhere between .5ohm and near 0. But when it goes up by 300x, you are going to have problems. And that's 30W in a single small spot. So a 30W bulb puts off how much heat? It isn't too much but it is also not insignificant. Concentrate that heat in an area of only a few square millimeters, something is going to melt.

The adapters that GN has with their 4090's are superior to those that we see failing.

Somebody at one of the suppliers fucked up, and these bad adapters got released somehow. It's likely a small number. If nVidia can release instructions on easily identifying the bad cables and sending out replacements, this can be quickly remedied. Other option, just replace all 100k adapters (100,000 is the last quantity of 4090's that have so far shipped, from what I have heard).
Either solution should be doable.
I have yet to see one where the connector on the GPU itself was damaged. Some where the plastic from the adapter melts and gets lodged in there have been seen, but the connector itself was ok. The plastic in the adapter is closest to the source of the heat, so that is what is melting.

If you have a 4090, just don't use the included adapter unless you have verified it matches construction of those that Gamers Nexus has (peel away the shroud and see if you can view the voltage rating on the wire, 300V so far has only been seen used in properly constructed adapters), or just get a different adapter.

So far, no "fires" have started, but you don't want to risk it. So be sure of the adapter quality, or use a different one. Keep in mind that adapters from unknown vendors (say something off Amazon) is yet another unknown quantity, you might get one that is of worse construction quality. Stick to a trusted brand, or go directly to your PSU vendor, or just wait for nVidia to finish their investigation.
 
Back
Top