NVIDIA GeForce RTX 4090/5090 Prototypes Tested with Quad 16-Pin 12V-2x6 Connectors

[AleksandarK]

Recent engineering prototype leaks reveal NVIDIA's power delivery design for its RTX 5090 flagship GPU featured a jaw-dropping quad 12V-2x6 connector configuration. The prototype board, while not publicly photographed, is significantly different from the final version we have today. The quad-connector design serves multiple purposes beyond raw power delivery. Each 12V-2x6 connector, capable of delivering up to 600 W individually, creates a theoretical power ceiling far beyond reasonable requirements. More likely, the implementation allows for separate voltage rails and enhanced power stability, crucial for the card's reported 575 W TDP - a 125 W increase from the RTX 4090's specifications. This overengineered piece follows NVIDIA's historical engineering practices, where early prototypes typically feature expanded power delivery systems that are later optimized for production.

The timing of this leak happens just in time with ongoing discussions about 12V-2x6 connector reliability, particularly following documented failures. While the quad-connector design may seem excessive, it distributes power load more evenly across multiple connection points, potentially addressing thermal concentration issues that plagued single-connector implementations. Board partners like ASUS have already implemented sophisticated current monitoring solutions in their custom designs, using shunt resistors to detect overcurrent conditions across individual pins. Established hardware leaker, Panzerlied confirmed the prototype's existence on Chinese forums, alongside images of other engineering samples featuring similar power delivery experiments across multiple GPU generations.



View at TechPowerUp Main Site | Source

 

[SAINT ENZO]

the transient spikes would probably be stable this way lol

 

[Crackong]

is significantly different from the final version we have today.

And I bet it is significantly SAFER (in terms of fire hazard) from the final version we have today.

 

[ymdhis]

There's a line, and we have crossed it.

 

[Hereticbar]

Four 12V-2x6 Connectors, you know what that means?

Quadruple the chance of "user error"!

​

 

[bonehead123]

Sooooo....

now we have gone from 1 to 2 to.... not 3, but 4 connectors just to temper the outrageous power draw of these new cards & keep them from m*e*l*t*i*n*g like a bucket of ice cream at the gates of hell....like, w.T.f..... ?

 

[AGlezB]

There's a line, and we have crossed it.

The cost of power is more power. We've been crossing lines from the moment the legacy ATX 20-pin cable became insufficient and we decided to put up to three 8-pin connectors on our mobos and up to four in our graphics cards. And we still want more power, at least 20% gen over gen, but we don't want 5-slot cards (4 is enough) and we don't like a bunch of cables blocking the LED ligthing so we wanted a thinner cable with a single small connector that will do the same job we used to do with 4 bigger and beefier ones... and then we act surprised when it has issues.

 

[Denver]

So that's what made the lights flicker... thanks, Jensen.

 

[hsew]

Having that many connectors kinda defeats the purpose of the 12VHPWR in the first place… just ditch it already!

 

[maxfly]

One step forward two steps back.

 

[LabRat 891]

Since, I can't imagine even their lab pushing 2.4+KW
I can only assume nVidia's engineers have been well-aware of the transient current issue, and did something else to mitigate Vdrooping.

 

[FoulOnWhite]

Naah, i do not want 4 cables from the GPU. More expense on cables, and PSU on a already expensive GPU. What about just 2 thicker wires, +12/-12. Works on a motorcycle battery that draws more amps than these cards. Wires don't have to be so thick, and with only 2 the connectors can be much beefier.

 

[Paganstomp]

( Creates a theoretical power ceiling far beyond reasonable requirements. ) Seem to be a paradox. What is a requirement for reasonable, anyway? 600 watts seems to be excessive for a home computer. That's the thought I have in the back of my head.

 

[Vayra86]

Four 12V-2x6 Connectors, you know what that means?

Quadruple the chance of "user error"!

​

Maybe its more like a plane at that point that can still fly on one engine. Maybe. For a while. The fun fact is you're not wrong. Also let's just make a mental picture of four sets of 12 ittybitty sleeved cables hanging off your GPU. Now don't you dare get those 48 cables twisted in any way, because you know, user error

 

[Solaris17]

So that's what made the lights flicker... thanks, Jensen.

That and laser printers.

 

[x4it3n]

4x 16-pins is insane and only for R&D purposes. I understand that Nvidia just want to use 1 cable to make their GPUs look nicer, but 2x 16-pin connectors on each RTX 5090 should have been mandatory since it would be a lot safer!

 

[freeagent]

They should just use a couple of XT60 connectors and call it a day. 600w is nothing in the RC world when you have lipos putting out serious amperage.

 

[x4it3n]

They should just use a couple of XT60 connectors and call it a day. 600w is nothing in the RC world when you have lipos putting out serious amperage.

It's not only the connector, the cable also melted! 20+ Amps is insane!

 

[randomTPUreader]

The GPUs need to take a page out of the car audio / RC world and step up to 8-12 AWG power cables with XT90 connections on either side. That should be good for 90-120 amps per cable at 12v.

 

[sagui]

4x 16-pins is insane and only for R&D purposes. I understand that Nvidia just want to use 1 cable to make their GPUs look nicer, but 2x 16-pin connectors on each RTX 5090 should have been mandatory since it would be a lot safer!

Considering the production board only had a SINGLE shunt mosfet for the entire cable, not even 2 connectors would make this thing safe, it would just buy you time before the cable and the PSU just gave up

 

[Crackong]

As buildzoid indicated.
These connectors are paralleled inside
and only being treated as one giant cable from the PSU and the card.
Therefore current flow through the path of least resistance
Hence the wire with better contact actually melts first.

The whole electrical solution is just flawed at the very beginning.

One more observation.
Noticed that both the burn and der8auser's hottest wire all come from the nearest pin from PCB

That pin has the shortest 90 degree bridge behind the connector, hence least amount of resistance.

 

[N/A]

That's the spirit a 1350 W TITAN with redundancy, but the final will probably only sport 2

Considering the production board only had a SINGLE

Those are stunt resistors. Not mosfets. I believe there are 3 here with individual controllers.

 

[Lycanwolfen]

Remember when Nvidia had that road map in 2016 where it was showing the future of GPU's less power and more performance in generations to come. Boy did they ever scrap that idea.

There it was suppose to be less power each gen. Boy they got that one wrong. According to that road map Blackwell would have be 100 watts maybe.

Like the Geforce GTX 480 to 580 was a drop of 100 watts but better performance same when it went from the 580 to 680 to 780 each time less wattage more performance.

Nvidia got better and better. Then they went complete reverse with the 2000 and 3000 and 4000 and now 5000 series more and more power hungry. Reminds me of the vacum cleaner card.

 

[ratirt]

Hmm. They really want to sell more brand new NV cards when the already purchased are damaged. What a plan

 

[WatchThe80s]

Let us just switch to the 230V PSU power cable for the GPU.