Nvidia RTX 4090 power connectors melting?

Add on to that, how does multiple rail power supplies respond when the different rails are shunted together in the connector? Need a power supply guru here.

From what I observed, previous GPUs with multiple pcie connectors had separate power circuity for each connector, while the new connector connects everything together so one power circuit in the card for the pcie connector. The 3090Ti didn't have this issue, so maybe a mute point. Unless the connector was split per 8 pin cable and not tied together like on the 4090. I don't know.
3090Ti have the 12VHPWR internally split into 3 separate rails. The input section on all I've seen PCB images of looks basically the same as 3x8-pin cards and from what my MSI Gaming-X reports to HWiNFO / GPUz it's treating the power input as if it is 3 separate 8-pins with decent balancing.

FE 3000-series have the same setup. My 3070Ti FE shows in software the "NVidia 12" or whatever we were supposed to call it as being 2x 8-pin and the input circuitry is duplicated like a native 2x8.

Lack of balancing on 4090 def seems like it could be part of the problem. Can't speak to how multi-rail PSUs would handle... Are true multi-rail PSUs still a thing? I haven't thought about those much since multi-socket SMP and SLI were the big thing in the 00s...
 
  • Like
Reactions: noko
like this
...how does multiple rail power supplies respond when the different rails are shunted together in the connector?..
The multiple rails are just have separate current limits applied, they aren't separately regulated. Thus, nothing bad will happen when they are connected together.
 
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

View attachment 524736
Compared to my Suprim X, which basically same card:
View attachment 524737
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/
View attachment 524739
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/
View attachment 524745
Color me shocked that people with melting connectors are inserting them wrong.
My Toyota Prius never had runaway acceleration either... guess that was just a myth too because it didn't happen to me.
That turned out to be a user error, just like this whole thing seems to be.
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.
The boards are not massive. This is a 2080 Ti FE (top) lined up with the 4090 FE and to-scale with the PCI-E fingers.

1668436646087.png
 
Better scrap the 8-pin connector, too.

3090:
View attachment 526666

3080:
View attachment 526668
Statistics are a thing. The 8-pin connector has been around for almost 20 years now, of course there will be incidents. The new connector barely came out and there are already a number of incidents.

Also the 3000 series GPUs consume a ton of power especially overclocked. Who knows what sort of crazy overclock those people were trying to do and how many amps were flowing through those cables. That's why many models have 3 8-pin connectors, not 2. It's easy to go beyond spec with an overclocked 3090.
 
Faulty 8 pin? Sure. Out of the gazillion of those there will be ones that melted etc. However, it's not even remotely comparable to the terrible new connector and adapter with the 4090. It's been a month and there are 100 cases. So don't even bring melting 8 pins into the equation.

You know what this 4090 situation is more akin to? Let's say there's a PSU released a month ago and there are about 100 cases of it's power connectors failing. You know what every single reviewer not on their payroll would say? Disgrace, disaster, safety hazard, dont use bla bla. Because that's how it should be. This situation with melting cables is exactly that.
 
https://smile.amazon.com/gp/product/B0BF4QBYV4

3x8pin to 16 pin 12VHPWR for a couple different modular power supply plug configurations. Make sure the cable you order is compatible with your power supply.

The one marked Corsair compatible looks to match both my Corsair RM1000 (bought in 2016) and Cooler Master V850 SFX.
 
https://smile.amazon.com/gp/product/B0BF4QBYV4

3x8pin to 16 pin 12VHPWR for a couple different modular power supply plug configurations. Make sure the cable you order is compatible with your power supply.

The one marked Corsair compatible looks to match both my Corsair RM1000 (bought in 2016) and Cooler Master V850 SFX.
That is the most chinese looking cable ive ever seen lol

Take note "Do not compatible with the older pcie..."

Manufacturer Dongguanshi Cunpin lndustrial Co.,Ltd

There has to be better out there than that right?
 
Everything is manufactured in China.
My point was this one was also designed and marketed in china and might not be a good solution to a bad connector design/implementation that is currently the issue at hand.
 
There has to be better out there than that right?
Right now? Not finding it. Next month? Maybe.
My point was this one was also designed and marketed in china and might not be a good solution to a bad connector design/implementation that is currently the issue at hand.
The Nvidia adapter is, according to Igor, made by 2 manufacturers in Taiwan. How's that going for 4090 owners?

I'd rather Corsair had their cable in stock and Cooler Master also had one available, but that's not the case.
 
Can someone please give the cliff notes on this video? I can't watch it now.
Your holding it wrong.

My take is that 12VHPWR adapter is a terrible design for DIY users. That needs to be idiot-proof, and the design is far from idiot-proof. Also, seems to be the more the connected is put through mating cycles, the chances for failure go up.
 
Can someone please give the cliff notes on this video? I can't watch it now.
User error is still an issue, poor design to allow user error to be an issue, possibly contamination inside the plug, other stuff too I really only skimmed around as I don't have 30 minutes of my life to give up on a video card I'm probably never going to have.

edit: Kyle said it better right above me.
 
^ At hardforum we knew that a week ago, using user evidence even: https://hardforum.com/threads/nvidi...ectors-melting.2022862/page-7#post-1045496995. In the post I think I had to push connector 3 times to get full contact seen in screenshot for my Suprim X 4090, just felt way more pressure needed vs 8 pin connectors.
I don't think GN claimed to have uncovered anything truly new- they just took the time and resources / $$$ to do a thorough investigation with actual labs to confirm / debunk the theories already out there
 
Can someone please give the cliff notes on this video? I can't watch it now.

User error and foreign object debris are two of the main reasons why some GeForce RTX 4090 owners have ended up with melted connectors, according to a new in-depth investigation by GamersNexus’ Steve Burke

Burke offered additional insight through his reddit account, echoing the idea that improper seating is one of the main issues:

The two primary ones, except it’s sort of like a 2+1 set of issues — 2 related to seating, 1 related to FOD. The seating one seemed to most effectively trigger failures when combined as a bad, specific angle on the cable route (towards the ‘a’ in the NV logo, since they’re oriented differently on some cards) PLUS a poor mount. We had trouble forcing failures when it was just one or the other. The FOD one, as a note, could be debris deeper/not cleanable by the end user also. We saw some molded into the strain relief. But it could also be burrs and damage from the dimples, according to the third-party failure analysis lab we sent it to.

(oh, one other thing – the high power contributes as well, maybe being the reason this one is failing more often than we heard about 3090 Tis fail or something)



https://www.thefpsreview.com/2022/1...oreign-debris-according-to-new-investigation/
 
You have to lube it if you want full insertion.

Best bet is to pay attention. If you are not able, retract the sheath, apply dielectric lube and drive as Deep as one can.
My grandmother had similar advice for 16 bit IDE cards
 
I suggest people watch the video again. Even though debris could be a cause of this happening, not a single verifiable instance of a melt connector points to it being the cause. It seems that nearly all the melted connectors so far reported have been user error.
 
I suggest people watch the video again. Even though debris could be a cause of this happening, not a single verifiable instance of a melt connector points to it being the cause. It seems that nearly all the melted connectors so far reported have been user error.
You can't say this because the cables didn't go to you but in this video it's a combination of user error AND design being the likely culprits as its really easy to think your cable is properly seated when it's not.

The cables don't click like an 8 pin connector does and really if you're sending this much power though them they should. A cable being idiot proof should be apart of the design.
 
Last edited:
You can't say this because the cables didn't go to you but in this video it's a combination of user error AND design being the likely culprits as its really easy to think your cable is properly seated when it's not.

The cables don't click like an 8 pin connector does and really if you're sending this much power though them they should. A cable being idiot proof should be apart of the design.
The connectors for my jeep doors have a lever that you pull until it latches and a button pops out to show it is tight and locked. I see posts all the time of "after i put my doors on my door locks dont work" and you see in the photo that it isnt latched.

I do agree with what you are saying but also it wont erase the problem, only reduce the cases of it which is still good but not enough imo. A thermistor or thermal fuse at the connector on the board might be helpful to monitor and shutdown if it gets too hot, at least to help prevent a meltdown.
 
  • Like
Reactions: erek
like this
The cables don't click like an 8 pin connector does and really if you're sending this much power though them they should. A cable being idiot proof should be apart of the design.
So you press the latch closed with your finger, skip the tear-filled essay to reddit about how latches should always click closed by themselves or you are stuck and can't go any further, and get on with your life. And I don't mean you but the overall grievance brigading around this issue.

That's not to say the tolerance could've been better on the connector so it auto clicks - everyone likes an auto click - but double checking during a build never hurts- "is my memory seated correctly? Is my cpu heatsink flush? Did I actually get the PCIe card into the slot I have no visibility of?" etc
 
Last edited:
So you press the latch closed with your finger, skip the tear-filled essay to reddit about how latches should always click closed by themselves or you are stuck and can't go any further, and get on with your life. And I don't mean you but the overall grievance brigading around this issue.

That's not to say the tolerance could've been better on the connector so it auto clicks - everyone likes an auto click - but double checking during a build never hurts- "is my memory seated correctly? Is my cpu heatsink flush? Did I actually get the PCIe card into the slot I have no visibility of?" etc
Yup and when the memory isn't seated correctly the computer fails to boot and sometimes will shutdown too. It also doesn't catch fire aka it's idiot proof. That's literally the case in every scenario you listed, except this one.

It's actually kind of strange it wasn't done here. Plug standards don't change all that often.
 
Yup and when the memory isn't seated correctly the computer fails to boot and sometimes will shutdown too. It also doesn't catch fire aka it's idiot proof. That's literally the case in every scenario you listed, except this one.

It's actually kind of strange it wasn't done here. Plug standards don't change all that often.
It evolves over time. CPU's didnt have an odd pin at one point either and you could put them in the socket wrong and fry them, makes a nice bright spark too.
 
You can't say this because the cables didn't go to you but in this video it's a combination of user error AND design being the likely culprits as its really easy to think your cable is properly seated when it's not.

The cables don't click like an 8 pin connector does and really if you're sending this much power though them they should. A cable being idiot proof should be apart of the design.
Let's call it a design-induced user error, then.
 
Last edited:
You can't say this because the cables didn't go to you but in this video it's a combination of user error AND design being the likely culprits as its really easy to think your cable is properly seated when it's not.

The cables don't click like an 8 pin connector does and really if you're sending this much power though them they should. A cable being idiot proof should be apart of the design.
Let's call it a design-induced user error, then.

To draw parallels to the medical device industry again, we are required to design products in such a way that they are not likely to result in user errors.

If the design is such that it is easy to misinterpret or get a bad grip and slip or anything like that, it is on us, the designers. We are at fault.

It is not sufficient to just put a warning in documentation. As part of usability engineering we are required to do a formative usability study with early prototypes or other stand-ins on actual representative customers, figure out where misunderstanding or other usability errors are likely, and then implement design changes to as much as technically possible (without regard for cost) eliminate them, and then prove they have been satisfactorily removed through a summative usability study.

Documented warnings and labels can only be used as a mitigation if we prove they actually work by testing on users.

Additionally we are also responsible for reasonably foreseeable intentional misuse by the end user, and must design our products to also mitigate intentional misuses.

Now granted, the stakes are usually higher with a medical procduct. Life and limb may be under threat, but that could also be said if a 4090 adapter burns down your house.
 
Last edited:
Thats great for medical devices where the consequences of a mistake are life threatening.

But that kind of scrutiny, testing and redesign takes time and money. If the computer industry adopted that burden of regulation, we would still be using 60Mhz Pentium1s with 16 color video cards.
 
Wouldn’t be hard for them to do a resistance test.. probably would have flagged the issue earlier.
 
Thats great for medical devices where the consequences of a mistake are life threatening.

But that kind of scrutiny, testing and redesign takes time and money. If the computer industry adopted that burden of regulation, we would still be using 60Mhz Pentium1s with 16 color video cards.
We're talking a power connector here. This isn't rocket science / medical grade. Anyone who designed this connector and thought it would be fine for the masses is an idiot. There is absolutely no reason they couldn't have just made this new 12vhpwr connector at least have the pin size of the standard 6/8-pin PCI-E connectors. It wouldn't take that much more room and would largely eliminate a lot of these issues.

I'm surprised major OEM's like Dell allowed nvidia to push this standard, TBH. It's the OEM's that are going to have to design their own solutions to fix this problem when it comes to shipping/assembly, and it's going to cost them more money. I'm curious to see how Dell is going to deal with this with their in-house 4080/4090's and what is going to happen to the case structure.
 
We're talking a power connector here.
Nah. Too difficult for the cutting edge in consumer electronics to handle apparently. No wonder we get treated like dogs begging for table scraps when we'll make excuses for something like this.
 
Nah. Too difficult for the cutting edge in consumer electronics to handle apparently. No wonder we get treated like dogs begging for table scraps when we'll make excuses for something like this.
I'm amazed at the rehashed Apple tactic of "You're holding it wrong" being applied to this as well. If a larger than normal portion of the population is doing it wrong then it's designed insufficiently.
 
Back
Top