https://www.techpowerup.com/300229/psa-dont-just-arm-wrestle-with-16-pin-12vhpwr-for-cable-management-it-will-burn-up
 
Despite sticking with PCI-Express Gen 4 as its host interface, the NVIDIA GeForce RTX 4090 "Ada" graphics card standardizes the new 12+4 pin ATX 12VHPWR power connector, even across custom-designs by NVIDIA's add-in card (AIC) partners. This tiny connector is capable of delivering 600 W of power continuously, and briefly take 200% excursions (spikes). Normally, it should make your life easier as it condenses multiple 8-pin PCIe power connectors into one neat little connector; but in reality the connector is proving to be quite impractical. For starters, most custom RTX 4090 graphics cards have their PCBs being only two-thirds of the actual card length, which puts the power connector closer to the middle of the graphics card, making it aesthetically unappealing, but then there's a bigger problem, as uncovered by Buildzoid of Actually Hardcore Overclocking, an expert with PC hardware power-delivery designs.

CableMod, a company that specializes in custom modular-PSU cables targeting the case-modding community and PC enthusiasts, has designed a custom 12VHPWR cable that plugs into multiple 12 V output points on a modular PSU, converting them to a 16-pin 12VHPWR. It comes with a pretty exhaustive set of dos and don'ts; the latter are more relevant: apparently, you should not try to arm-wrestle with an 12VHPWR connector: do not attempt to bend the cable horizontally or vertically close to the connector, but leave a distance of at least 3.5 cm (1.37-inch). This ensures reduced pressure on the contacts in the connector. Combine this with the already tall RTX 4090 graphics cards, and you have yourself a power connector that's impractical for most standard-width mid-tower cases (chassis), with no room for cable-management. Attempting to "wrestle" with the connector, and somehow bending it to your desired shape, will cause improper contacts, which pose a fire-hazard.
 
The 12VHPWR connector is a new standard, which means most PSUs in the market lack it, much in the same way as PSUs some 17 years ago lacked PCIe power connectors; and graphics cards included 4-pin Molex-to-PCIe adapters. NVIDIA probably figured out early on when implementing this connector that it cannot rely on adapters by AICs or PSU vendors to perform reliably (i.e. not cause problems with their graphics cards, resulting in a flood of RMAs); and so took it upon itself to design an adapter that converts 8-pin PCIe connectors to a 12VHPWR, which all AICs are required to include with their custom-design RTX 4090 cards. This adapter is rightfully overengineered by NVIDIA to be as reliable as possible, and NVIDIA even includes a rather short service-span of 30 connections and disconnections; before the contacts of the adapter begin to wear out and become unreliable. The only problem with NVIDIA's adapter is that it is ugly, and ruins the aesthetics of the otherwise brilliant RTX 4090 custom designs; which means a market is created for custom adapters.
 
I think it is sad that this information is coming out so late after the launch of the RTX 4090. 
 

Guess they need to rethink/improve this "power solution" 
 
I'm left wondering how they ever made it through validation with this socket/power design, without these restrictive limits being known? 

 
(excerpt from OP link)
"NVIDIA even includes a rather short service-span of 30 connections and disconnections; before the contacts of the adapter begin to wear out and become unreliable."
 
 
I think those photos (found in OP's link, Update) of the semi melted Nvidia 12+4 pin adapter & GPU socket were impressive - & show the results of loose connections & the heat generated very clearly
 
(excerpt)  "Update 15:59 UTC: A user on Reddit has a report with pictures how his 16-pin adapter burnt up."
 
No mention if this person had modified the BIOS, how the wires sere routed (bent?), cycled the connector many times or how the GPU was being used before the failure
 

Cool GTX
(excerpt)  "Update 15:59 UTC: A user on Reddit has a report with pictures how his 16-pin adapter burnt up."
 
No mention if this person had modified the BIOS, how the wires sere routed (bent?), cycled the connector many times or how the GPU was being used before the failure
 

Yea there is some information missing in the recent reddit posts.  Certainly if they were using a modified BIOS that could pull more power than the cable is spec'd for, it could cause that problem.
 
I think the wires getting pulled from their terminals on those tight bends (causing an electrical arc) is the most logical reasons to me.  Glad that Cablemod identified this so early on, and it is a shame that this didn't all come to light with all the testing/RnD resources at Nvidia's disposal.

Killmur
With this recent development and the fact that newer Intel\AMD CPUs are starting to draw more power than necessary I think it's high time the entire power delivery system to the motherboard, cpu and gpu is redesigned. The 22+ year old industry standard 24 pin connector needs to be redesigned to provide high enough power so we don't have to worry about things like this anymore.

I mean, this is the reason that Intel has a separate, and more preferred ATX12VO spec (at least to Intel) and prefers that the industry begin moving to that, rather than continuing to use ATX standards.  ATX12VO 2.0 was also recently published, and if you're not familiar with ATX12VO, it does essentially change the entire power delivery system to everything connected to the motherboard. 

Killmur

EVGATech_LeeM

Killmur
With this recent development and the fact that newer Intel\AMD CPUs are starting to draw more power than necessary I think it's high time the entire power delivery system to the motherboard, cpu and gpu is redesigned. The 22+ year old industry standard 24 pin connector needs to be redesigned to provide high enough power so we don't have to worry about things like this anymore.

I mean, this is the reason that Intel has a separate, and more preferred ATX12VO spec (at least to Intel) and prefers that the industry begin moving to that, rather than continuing to use ATX standards.  ATX12VO 2.0 was also recently published, and if you're not familiar with ATX12VO, it does essentially change the entire power delivery system to everything connected to the motherboard. 

That's good to hear then cause we gotta stop relying on so many extra power cables which in turn causes extra issues. I remember when 4-pin was standard for base line cpu usage and now it's 8-pin and the same has been happening to the gpu side.

The issue is that with ATX12VO you need a 12V-only power supply, which turns your motherboard into sort of a DC-DC converter and handles any 3.3V/5V connections for components, which also means the power connectors to them will come straight off the board.  It's not so much that it's a problem, but it means your PSU, MB, and all the other components are spec'd to work with that kind of an ecosystem.  Ultimately, systems can become much, much more energy efficient (and this is Intel's primary driving interest here), but it takes a large buy-in from all the hardware manufacturers.  These days, we use a lot fewer SATA and 4pin Molex connectors than we used to, so this is a lot more feasible than even 5-10 years ago. 
 
I will note that my explanation here is probably a bit simplistic, but this is the general idea of how a 12VO system would work.  It may be possible to also have an ATX12VO PSU that could provide auxiliary connectors to 3.3V/5V devices or power directly to VGA cards for flexibility, so don't look at my description above as a literal picture of how things have to work.  But systems would look a lot different, for sure.

^^^ Yes

this (thread) looks like a good explanation of the root cause of the issue with the NVIDIA adapters
 
 
The horror has a face – NVIDIA’s hot 12VHPWR adapter for the GeForce RTX 4090

the_Scarlet_one
It doesn’t make the buck converters efficient, it make the PSU itself efficient, and make motherboard manufacturers figure out how to manage the other power ranges they need. Maybe everyone will move to 12 volt, and kick 3.3v and 5v to the curb completely.

It also gives MB makers an excuse to increase costs exponentially to accommodate the new board design requirements.  Feature rich boards are already $400-$700.

808sting
 
It also gives MB makers an excuse to increase costs exponentially to accommodate the new board design requirements.  Feature rich boards are already $400-$700.

True  +1

Kinda tired seeing pricey boards.  I know it will become the 'norm' for starter boards some day.  

808sting

the_Scarlet_one
It doesn’t make the buck converters efficient, it make the PSU itself efficient, and make motherboard manufacturers figure out how to manage the other power ranges they need. Maybe everyone will move to 12 volt, and kick 3.3v and 5v to the curb completely.

It also gives MB makers an excuse to increase costs exponentially to accommodate the new board design requirements.  Feature rich boards are already $400-$700.

When you make a board for people already buying a $1500 GPU you can charge a premium since money is no option to those buyers or at least some of them.

Hoggle

808sting

the_Scarlet_one
It doesn’t make the buck converters efficient, it make the PSU itself efficient, and make motherboard manufacturers figure out how to manage the other power ranges they need. Maybe everyone will move to 12 volt, and kick 3.3v and 5v to the curb completely.

It also gives MB makers an excuse to increase costs exponentially to accommodate the new board design requirements.  Feature rich boards are already $400-$700.

When you make a board for people already buying a $1500 GPU you can charge a premium since money is no option to those buyers or at least some of them.

True.  I still work for a living, but am fortunate to be able to enjoy hobbies.  I can't understand those that can afford it, but aren't willing to pay for the best practice options.  Buying a 4090 and get a cheap minimum required PS is hard to understand.  Or make a 1000HP car and not upgrade the rest of the stock drivetrain.

ty_ger07

808sting
I can't understand those that can afford it, but aren't willing to pay for the best practice options.

If it is the minimum required, it should work.
If not, increase the minimums.

Correct and in my haste, I didn't phrase the context well. 
 
When paying so much for a high-end halo product and you have the funds, why not factor in some headroom and scalability in the system.  In the end, it comes down to personal resources and choice.  I see it all the time where MQs get you into trouble very quickly.  Majority of the times, change orders drive up the cost due to questionable decisions.
 
nVidia's supplier for the 16-pin adapter did a poor job at design and the one bad QC at Igor's lab based on the few cut open samples floating around.  A poor choice for the flagship halo product.

I have been thinking about Igor and Jayz explanation, and how they say the cracked or disconnected solder likely causes heat, but if the cracked solder is the cause wouldn’t the melt happen near the soldering, not near the male-to-female pin connection.

I think Steve is on to something with the fact the issue isn’t occurring where the solder joint is, but that design is still garbage compared to having the wires crimped to the pin, and then solder if they want more contact.

I hope Steve and his team look further into this. They seem to really do justice to the testing versus some others.

There's a post on reddit from a user showing there's a blob of nylon plastic obstructing the interior pin. If plastic got into the connector during manufacture that could very well explain the entire mess https://i.imgur.com/vEP0DjU.jpg

QuintLeo
The problem here is that the "no bend" guidance IS NOT POSSIBLE with the 4090 in many or most cases - you HAVE to bend the cable pretty close to the connector to make it fit AT ALL in the case.
Didn't have that issue with the old 6-pin and 8-pin PCI-E power connectors.
 
Best to avoid ANY card that uses these connectors, given all the the "burn-up" issues and BAD DESIGN on this piece of junk connector.

Cases?!?! Pfft. Don't need no cases. Just get a test bench. Problem solved.