Small nitpick: this means you can find out one or the other but not both. I think you meant they're orthogonal, which effectively means "unrelated" or "independent". Wikipedia says "When performing statistical analysis, independent variables that affect a particular dependent variable are said to be orthogonal if they are uncorrelated".
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
-
Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.
RTX 5090 FE Molten 12VHPWR
- Thread starter erek
- Start date
littlegreenmenfromhome
Limp Gawd
- Joined
- Aug 27, 2025
- Messages
- 385
No, you keep ignoringmubers and pretending user posts on forums are data in suffiecient amount.
Perhaps posting the data again will work this time:
https://www.reddit.com/r/nvidia/comments/yyqffy/nvidia_responds_to_melting_cables_warranty/
- Nvidia Official Statements:
- "We are actively investigating the reports. We are aware of about 50 cases globally. Our findings to date suggest that a common issue is that connectors are not fully plugged into the graphics card. To help ensure the connector is secure we recommend plugging the power dongle into the graphics card first to ensure it's firmly and evenly plugged in, before plugging the graphics card into the motherboard."
- "We are investigating additional ways to ensure that the connector is secure before powering on the graphics card. NVIDIA and our partners are committed to supporting our customers and ensuring an expedited RMA process, regardless of the cable or card used"
- "Nvidia has been able to test the cables that were RMA'ed by affected customers. In all of the cases a wear line is clearly visible that indicates the cable wasn't fully inserted into the 16-pin power connector"
- "Anybody who has an issue [relating to this] will be taken care of. We'll expedite an RMA"
- "Any issues with the burned cable or GPU, regardless of cable or GPU, it will be processed"
- Nvidia Official Article: https://nvidia.custhelp.com/app/answers/detail/a_id/5413
- GN Updated research:
- 0.04% to 0.05% failure rate range
- Sales source = 125,000 units. Updated GPU sales via 4 board partners
- Connector burn count source = 50 units. NVIDIA statement
- 50/125000 = 0.04% failure rates
- GN again showed that the cable could look like they are inserted but not fully seated and you can wiggle it out. This means you need to push the connector more to fully seat it.
- GN referred to Tomshardware article talking about a paragraph within the warranty manual that stated it "may" void manufacturer warranty.
- GN Confirmed with Nvidia that this is not the case and Nvidia will honor the warranty
- GN asked Nvidia about their foreign object debris findings but no confirmation on this side
- How to improve the connector
- GN thinks latching mechanism can be improved
- Changing the sense pin length where if the connector is not fully seated, the card won't turn on
- GN showed a way to properly install the cable
- Please watch the video:
View: https://youtu.be/_QmKYJzJhB4?si=qOo7QAwsh6_9JoLU
- Please watch the video:
This is a far larger issue, but data was also provied here, yet people are silent about that:
https://www.pcguide.com/news/nvidia...defective-rtx-50-series-gpu-from-their-store/
Now keep trying to handwaive the data away
Your "data" is from four years ago. Also, the fact that you believe a company is being honest and not sweeping the numbers under the rug is concerning. Nobody will ever know the true extent of how many 12VHPWR connectors and cable have failed but I do trust the guy from Northridge Fix who was contracted by nVidia to fix them and he has stated that the numbers were extremely high compared to other repairs that he has done.
Yep that guy is getting blocked as he's annoying and doesn't listen to anything unless Nvidia God says so. While anyone with 1/2 a brain can look at the thing and realize it's junk. Also you can't go by the total number of cards as we know it's safe to probably 450-500w so really only the 4090 and 5090s failure rate matters.
sharknice
Supreme [H]ardness
- Joined
- Nov 12, 2012
- Messages
- 5,016
- Joined
- Dec 2, 2004
- Messages
- 32,800
Nvidia's statement was about 4090s.
It’s from 2022. I wonder if any have failed in the meantime? Maybe it’s up to 51?
- Joined
- Dec 2, 2004
- Messages
- 32,800
Possibly. It's a very rare defect from what we've seen on forums and reddit over time. If they were melting en masse like people here seem to think, we'd be seeing dozens of complaints a day. Nvidia would be forced into issuing a recall if the connector were really that bad. Instead we see one or two reports every so often, and they get sensationalized by you tubers and forum fanboys, despite both amd and Nvidia using the connector.
Last edited:
Probably after a hefty Furmark session...
It really does seem like too much is made out of this. Not that we shouldn't want a more robust connector but that people act like this happens all the time, or that it is inevitable that everyone who has a GPU with one will have it melt. That's just not the case. Something else to recall is that these are used in datacenter GPUs as well. You know, all that shit doing all the AI that people use to make AI slop mocking the connector. If it were a big melting problem, it would be a big problem there (since those get run hard) and that would be something they'd have to deal with. Those companies that are dropping $10k or more per RTX 6000 and buying hundreds or thousands of them are not going to sit around and be ok with it if they melt all the time.
And while it is another sample size of one, I'll again point to myself: I bought a 4090 3 years ago, it of course used 12VHPWR. Used it in my system for those 3 years with the Seasonic cable for my PSU (it is not a native 12VHPWR PSU) and when I bought my 5090, there was no damage. I even posted pictured on Hardforum. Of course that's not news, and heck the only reason I even bothered to make a post was because people get so worked up over the connector.
I would like to see a better design because I think high safety margins and high tolerance to user error are good things for consumer products. However people need to stop taking every melting connector story as some major deal and proof that it happens all the time. It doesn't, it happens on occasion, usually because of a confounding factor (hard cable bends often being the issue) and has happened long before 12VHPWR was a thing. Search around and you can find old 4-pin Molex connectors that burnt. It just wasn't considered very newsworthy.
StoleMyOwnCar
2[H]4U
- Joined
- Sep 30, 2013
- Messages
- 3,511
https://old.reddit.com/r/pcmasterra..._all_know_the_12vhpwr_is_a_bad_connector_but/
This has been brought up before, and I think this topic debunks it fairly well with plenty of potential reasons. Namely,
- even if there was an issue, the warranty is so advanced that people with these wouldn't even notice an issue.
- they wouldn't be posting it anywhere, these are completely different consumers
- it's highly likely that they use a different means of powering them to begin with
- they're nothing like consumer gpus at all
So no, datacenter GPUs are a completely different beast and aren't in the scope when talking about this issue at all whatsoever. And for the ones that have the 12VHPWR connector, yes, they do fail. They actually fail a lot if the posts in that topic are anything to go by.
The problem with this is that again you're just basing this on some static number. That's nonsense. Failure rates for this cable should be rated in terms of "MTTF", or Mean Time To Failure (or alternatively, MTBF, since HDDs can be replaced I guess, and so can these). Why? Because it basically should use the exact same model platter hard drives use. Why? Because that's the problem with hard drives. Even if they were built perfectly, if any little thing goes out of spec, the HDD can go crashing down. It's subject to physics and thus random physical chaos. This cable is exactly the same. It's theoretically """enough""" for the loads that it's delivering, but I'm not sure how much it will take to get you to understand that it's not enough for "loads+random physical circumstances happening". Like gradual shifting in the cable and the pins over time due to settling, or you picking up and putting your case down, any of these can eventually cause it to just fuck up and die because it DOESN'T HAVE ENOUGH SAFETY MARGIN FOR THAT, not when it's running at its max rated wattage. No one knows how much it will take, it depends on the exact card it's connected to and a multitude of other factors. These GPUs can on average keep running fine for years. But eventually if any thing goes wrong, it can easily melt.
That's completely different from an 8 pin connector, where if the design on both sides of the 8 pin connector was perfect and had no flaws, they're not really subject to physics. It doesn't matter. They have so much margin of error that the connector will pretty much never fail unless a component on either side of it does, too. Like this is how theory works when you're building anything that's going out into the wild. Have you ever played a bridge simulator game? Well you should know that you're going to make tons of shit in that game that technically scrapes by with the minimum budget use possible but there's no way in hell you release that out to the public. Because that's what theory is about. Theory is the baseline, and then you do baseline++ to make the actual production version. You have to expect factors outside of your control will go wrong.
So I don't know. Maybe the "failure rate" will remain low, but I want to point out that like others have said, you're just taking a party at their word, one which has a kind of fucking massive vested interest, and that also this is frankly the wrong type of failure scheme anyway. Overall though the main issue is that we have way too little credible information, so I'm going to lean on the fact. Not theory, but actual irrefutable fact, that these cables have practically 0 margin of error when driving the 5090 especially.
Last edited:
Depends entirely on your vendor. Dell servers? Ya pretty good. Exxact? Not as good.
You don't think anyone who posts on Reddit or Hardforum works with high end hardware? I don't work for an AI company, but I do work for a university that has servers with these things in them and I'm one of the people responsible for managing the servers, including dealing with warranty repairs. The people who run data centers are not orthogonal to people who play videogames.
They use 12vhpwr. Go look on PNY's site if you want to see, they have a diagram of the power connector.
Ya they are, they are the same basic thing but bigger. That's how nVidia was able to get in to this market so seamlessly: They didn't have to make a new product, what they already had worked great.
So your issue is there's too little credible evidence these connectors are good... and you are basing that on people on the internet screaming they are bad? Because that's good evidence? Sorry man but the fact that you see the occasional person bitching about it, particularly techtubers who have a reason to make hay out of this, does not indicate to me widespread problems. I'll need better evidence that there is a big problem before I believe it.
Because the Internet is known to make mountains out of molehills. An example I can think of recently was supposed issues with Phison controlled SSD drives and an Windows update. Supposedly a Windows update caused these things to start blowing up. I saw it numerous places on the Internet, but nobody had any evidence other than "this one guy said. Phison and MS said they tested and found nothing. We didn't see anything at work, no increases in failures. So what's more likely: It was some huge coverup by MS and Phison, both who do have an interest in it not being a problem and I (and others) who manage large orgs got lucky... OR someone online had an issue, made a connection that didn't exist, and the Internet blew it up bigger than it should be?
Again, I am NOT saying it doesn't happen and I am also NOT saying that we shouldn't want a better designed connector. What I am saying is that the amount of screaming seems to be way out of proportion to the actual issue. Heck, ask around on Hardforum or take a poll. We have a lot of nuts here who have and do own 90 series GPUs. See how many have had a melted connector, and see if there was anything else surrounding that (like the cablemod cables that weren't built properly, or a hard bend).
Look, like I said, I don’t need Nvidia’s data—we have plenty from the internet—and besides, there’s a whole new industry out there producing so many products just to try to solve this problem. Do you have anything similar for other cables? No?
Are there that many experts saying there’s a problem with 8-pin or other cables? No? Do you see any of these experts saying there’s a problem with the 8-pin connector? No?
So you can wait as long as you want for Nvidia to shoot itself in the foot, but that’s not going to happen.
I inserted a thermal sensor with display between the cables near the connector, as well as a fan to cool them, so that with enough airflow, the temperature is about 10–15 °C lower (from 45-50C).
For me, this is a better solution than a thermal camera—because I can see the temperature at any given moment
As an Amazon Associate, HardForum may earn from qualifying purchases.
The problem with all this is that the process takes time—first the connector (the pins) has to degrade, and then it has to melt, because as degradation increases, the temperature rises, and that’s just basic physics.
Nvidia created the problem on purpose, so it’s really funny that they’re going to “investigate” what it is.
Last edited:
- Joined
- Jul 17, 2006
- Messages
- 14,411
I wouldn't say the failure rate is on purpose, it likely has to do with contractual obligations for the connector supply/manufacture. They wanted a smaller connector and got one. However the mating surfaces appear to have poor surface area contact and result in high resistance connections in many instances. This leads to heat-up due to the high current, and results in melting of the plastic among other damage.
I have, as an engineer in Aerospace, had to explain this to people so many times over the decades. They go on and on about how much better something is, or how much safer or secure it would be to do it the way they've designed it. But, of course, it involves more steps or more hassle from a user point of view. My response is always "the more complicated you make it the less likely the user will do it that way". The response to me is always "then that's their fault!" .... *sigh*. My answer is simple ... an engineer's job is to make it as easy to use and still make it acceptably safe. Yes, sometimes people will ignore all attempts at safety and we can't stop that. But, if we make it a hassle to use safely for even the most sensible user then we're guaranteeing our work will be used unsafely.
We can't fix stupid but we can at least make it relatively pain free for someone who's interested in using things correctly.
StoleMyOwnCar
2[H]4U
- Joined
- Sep 30, 2013
- Messages
- 3,511
Has what things in them? I don't see a university having nearly as much funding as a biotech company. Nor do I know how much you actually push them normally.
I highly doubt every actual enterprise variant is using 12VHPWR. The post I quoted says that the ones in the picture are actually a rare variant for enterprise. Furthermore... what you have to realize is that at the scale they operate at, enterprise can just scale back the wattage. Or the vendor might do it for them, if they notice the failure rate is too high.
No they're not. They could for instance actually build per-pin rate limiting into enterprise PCB, and/or just not tie every pin together like our cheapo consumer boards do. Have you actually inspected the PCB? You're coming into this with a lot of assumptions and not much data, despite seemingly wanting to lean on data.
Okay this is the part that's starting to get annoying. It's like you guys just keep literally skipping over the specs sheet of the thing you're vehemently defending. Like just do a 5 minute google search. Look at Wikipedia.
https://en.wikipedia.org/wiki/12VHPWR
Let's briefly drill this into heads: THE CABLE IS LITERALLY RATED FOR 600W MAX. If BARELY A TINY BIT ABOVE THAT. Like I don't know how to make this more clear. That's what all of these youtubers, these theorists, and literally the flipping docs on the cable itself will tell you.
We aren't discussing numbers outside of literal objective specs sheets, corroborated by empirical testing. Running at 600W, this cable has a measly 1.14 safety rating. And this is made EVEN WORSE on consumer models specifically because the actual implementation on the GPU side has ALL OF THE PINS TIED TOGETHER. Any one pin goes wrong, the entire thing possibly just fucking dies. Also, let's talk about the instantaneous power spikes?
I'm going to fall back to the bridge simulator game example and point out that this is in the same vein as making a fucking bridge made out of hardened paper because "well, it passed my 3 simulations" and shoving it out in the real world, and then having a surprised Pikachu face when a thunderstorm happens and it just melts into rubbish. And it was actually a big enough issue to where Nvidia has had to make statements and whatnot (well they don't give a shit about us, so they were quick to blame it on user error).
Why aren't they failing more? Probably because unlike bridges, connectors are rarely going to be moved. Most people aren't going out and buying 3-4k GPUs, put into 5-7k+ systems, and then just tossing them around. Plus, most consumers are frankly probably wary of the thing anyway, if they're spending this much. Like, my HP prebuilt very carefully angled the connector down to minimize stress on it and keep it from bending too soon. OEMs are being ESPECIALLY gentle with them.
What I need to drill into your head, though, is that this cable failing isn't really a matter of if, it's a matter of when, or rather, what circumstances. If you treat it gently, adhere to the clearance bending guidelines, and even undervolt the GPU... it could basically easily last the life of the card. So I'm not trying to sensationalize. It can definitely last. But you have to be careful with it. That's what makes it a shit design. It has no safety margin. Failures can and will keep happening on it, especially with 5090s and to a lesser (much lesser, considering the wattage chart vs safety factor chart above) extent 4090s. A cable shouldn't just fucking melt because "welp they didn't insert it all the way lol". That's just garbage design, and all of the sense pin patches were just retarded bandaids over a dumb issue of "we just designed this poorly".
In professional systems, I highly doubt these connectors will be moved much or touched at all after setup. They probably aren't going anywhere. And hell I wouldn't be surprised if the cards were just undervolted. And they might have PCB patches to help. In a common consumer system, it's much more likely that they will, while having to mess with fans, read bios codes, or just their case being too tight and/or them vertically mounting their GPU. In these cases, there will be wear on the connector. Will it eventually fail? Maybe, maybe not. The fact that the possibility exists precludes it from being a good connector. It can never be a good connector. It's trash. It has no safety factor. Period. All of the theory points to it just being shit.
Anyway, I'm going to pull out the Mandalorian "I have spoken", because this is getting tiring and going in circles. Believe what you want to believe. I'll just power limit mine to 450-470 watts and be happy with my drastically increased safety factor. Thanks.
RTX 6000 Blackwell GPUs for doing AI research. Based on the amount of power they draw? They push them a lot.
Ahh the "I don't know so I'm going to assume that it is what supports my argument." Dude, the reason they developed it was for higher power density for things like servers. But ok, here's a picture of a cable from one of the Exxact servers:
Hmmm... What does that look like and where have I seen it before?
You are also doing a lot of assuming. You are correct I can't rule it out... but I can say it is nowhere in the spec sheets (yes I checked). The only board I find with any per-pin monitoring is the ASUS Astral.
You seem to be missing two things:
1) I have said, many times, I think we should have a better connector. My personal favored approach is to just add two more pin sets, have 8x2 pins and the same limits. That is still a small connector, but gives a much higher margin.
2) The fact that it is running close to the rated limit does not imply it will fail. This idea that it'll just melt after a certain amount of time is not accurate.
Two things can be true simultaneously: That the connector is not as good as it should be and that the reports of melting are highly overblown.
Blade-Runner
Supreme [H]ardness
- Joined
- Feb 25, 2013
- Messages
- 5,655
I never thought I would see the day where Ngreedia's retardation would spawn an entire new tier of overpriced PSUs and gadgets designed to stop GPUs from self combusting...
View: https://www.youtube.com/watch?v=jcp7OBD8XpU
View: https://www.youtube.com/watch?v=jcp7OBD8XpU
- Joined
- Mar 18, 2010
- Messages
- 14,448
Which are contradictory statements in terms of end goal and are no different than a politician playing both sides of the field. To get nVidia to change the design, this needs to be made as big a deal as possible. Saying reports are overblown implies nVidia doesn't need or can drag their feet on implementing change.
Blade-Runner
Supreme [H]ardness
- Joined
- Feb 25, 2013
- Messages
- 5,655
d3athf1sh
[H]ard|Gawd
- Joined
- Dec 16, 2015
- Messages
- 1,450
i made a post about this topic in the PSU section of the forum if anyone's interested. didn't want to anger the forum fairies. 
it can be viewed here 12VHPWR &12v 2x6 Fire Risk Discussion
it can be viewed here 12VHPWR &12v 2x6 Fire Risk Discussion
Blade-Runner
Supreme [H]ardness
- Joined
- Feb 25, 2013
- Messages
- 5,655
https://videocardz.com/newz/club386...unders-edition-was-taken-out-by-12v-2x6-cable
Weird how the thing which never happens keeps happening.
Weird how the thing which never happens keeps happening.
And it's going to keep happening the cable can't do 600w long term end of story. Honesty once I got the card in my hand and realized how shity the cable is I regreted not buying one with two plugs because now I have to limit my card to 450w
I don't limit my card; I've simply inserted a temperature sensor between the cables near the connector and added a fan to cool the connector.
If the temperature gets too high, I'll just replace the cable with a new one.
Problem is by the time it gets hot the female end might get damaged
Yes, but the female connector is more durable, so if I replace the cable regularly and keep them cool, both will last longer.
And I keep an eye on the wattage during games; most games run at around 350–450 W by default, so when you limit the graphics card to 400 W, the difference is really negligible.
The temperature sensor I installed, with cooling, reads about ~36°C at 400 W and 40°C at 575 W. Yes, the sensor is 3 cm away from the connector, meaning the temperature isn’t measured at the pins, but since the pins are soldered to the cable, the temperature can’t differ by much.
Last edited:
See mine is on water so in just about any new game the thing wants to pull 600w full time. I did the same thing temp probe and fan and while it stays cool I just don't trust it to not degrade and kill the card in a few years.
At higher temperatures, degradation simply occurs faster, so yes, cooling does extend the connection's life a bit, but the sensor will indicate when problems start to arise so we can inspect it and take the necessary action.
So this is all we can do in this situation.
Soldering both ends of the cable is best, but...
- Joined
- Jul 8, 2016
- Messages
- 471
i'll sum it up: 'thank you, leather jacket, et al.'
Blade-Runner
Supreme [H]ardness
- Joined
- Feb 25, 2013
- Messages
- 5,655
https://videocardz.com/newz/asrock-...-system-after-rtx-5090-power-connector-melted
Still happening even with new PSUs designed to shut down when the 12V-2×6 connector overheats.
Still happening even with new PSUs designed to shut down when the 12V-2×6 connector overheats.
I have the same psu. It's a gimmick they just put a temperature probe that's not even in the center of the plug....by the time it gets hot enough to trip the other pins of the connector could easily melt.
I hate to say it but I think in the end 100% of 5090s will die because the plug fails eventually. That's the problem with marginal electrical connections they degrade a bit each heat cycle and eventually it will get to the point it melts. Also changing the cable might help or make it worse as each time you remove/mess with the cards connector you are loosening the contact points a bit more.
I'm kinda thinking if you plan to keep a 5090 for a few years it might be worth it to find a good repair shop that can rip the plug off and solder wires right to the PCB or something similar
StoleMyOwnCar
2[H]4U
- Joined
- Sep 30, 2013
- Messages
- 3,511
Is this probe at both ends, or just at the GPU end? It looks like the GPU is operational, so maybe it did its job. The PSU end completely melted.
Anyway, that's why I limited my card to essentially 4090 TDP. At 450W, I think this thing will be fine. My 4090 has been operating nonstop 24/7 with zero issues at stock wattage, and I think 4090s have a much, much lower burn down rate in general. As far as everything else, the main thing to watch out for is the physical stressors on the connector. Follow regulations in terms of giving it the recommended amount of space before any bends, and treat the connector very carefully.
Like this isn't a joke. These things are burning down in real time. But, your PC can be a very controlled environment. In theory I think this connector can run safely. But operating at 100% of its spec? Yeah I don't think so. Time-based degradation will inevitably happen. I would recommend giving it no more than 500W, maybe 475. At that speed it should still be quite a bit faster than 4090 just due to the upgraded specs, and I think it should have much less chance of burning out over time.
The problem is that, like I said, this will KEEP happening over time. And we don't know the average failure rate or timespan between each failure. It could on average be fine for 2 years, 3 years, maybe even 4 years. The cards are too new to properly have real failure rate statistics. We'll be a few years out from now before we truly understand the impact of this.
Yeah it's only on the GPU end and once you plug the cable in the probe is about an inch form the plug and 100% pointless hints why in my picture you see I added two probes. One is inside the cable the other is right above the solder joints on the pcb.
I agree with you for 24/7 I limit it to about the same as with the water block I'm still slightly faster at 500w then it was at 600w stock heatsink out the box.
Last edited:
StoleMyOwnCar
2[H]4U
- Joined
- Sep 30, 2013
- Messages
- 3,511
Well that's why I had the idea in this topic:
https://hardforum.com/threads/what-...-to-a-relay-on-the-12vhpwr-connector.2048036/
Frankly this would be almost trivial to implement, too, compared to all of this proprietary cable nonsense.
ZeroBarrier
[H]ard|Gawd
- Joined
- Mar 19, 2011
- Messages
- 1,474
At some point even members of this forum have to admit that the issue is users not plugging the connector in correctly. I still have faith that a PC enthusiast forum will eventually see reason, but the world in general is making me think my faith is misplaced.
Article says, "the cable was fully seated at both ends and had no sharp bends" and then proceeds to provide a picture where the connector is not fully seated.
StoleMyOwnCar
2[H]4U
- Joined
- Sep 30, 2013
- Messages
- 3,511
Yes, a cable that looks like it's basically been completely melted internally at this point... isn't fully seated. And you're just assuming that they didn't try to pull it out or something to inspect it, and that's why that gap is there. Nope, that's just exactly the way it was. The design is absolutely flawless and it's 100% user error.
Jesus fucking christ.
Like I don't even care if it wasn't originally fully seated. If you have this sort of failure mode for "oh shit we got a 1mm gap here", then you fucked up designing the cable. Period.
Jesus fucking christ.
Like I don't even care if it wasn't originally fully seated. If you have this sort of failure mode for "oh shit we got a 1mm gap here", then you fucked up designing the cable. Period.