Unofficial 12V-2x6V Power Connector Melts NVIDIA GeForce RTX 5090

[Crackong]

Der8auer got the card & cable & PSU.
Observed and compared the results with his own 5090.
Finds out there are 2 wires carrying 20A in stress testing. (Other wires carried ~5A)
It makes those 2 wires extremely hot and dangerous.

 

[PixelTech]

This does seem like the interesting part in this case. While burning 4090 GPUs seem to have been common enough I do not remember that many pictures of the same thing burning on PSU side.

This is the PSU used:

ROG Loki SFX-L 1000W Platinum | Power Supply Units | ROG Global

ROG Loki 1000W Platinum is a high-wattage PSU for boundary-breaking SFF builds.

rog.asus.com

It only says PCI-E 16-pin but also ATX 3.1 that would imply 12V-2x6.

Also, as the PSU already implied, this is an SFF build.

A4-H2O - LIAN LI is a Leading Provider of PC Cases | Computer Cases

A4-H20A TRIBUTE TO SFF The A4-H2O is an 11-liter small form factor case, a collaboration project with DAN Cases. Based on the original DAN A4-SFX case sandwich layout concept, the A4-H2O continues the minimalist and compact form factor with improved hardware compatibility (including AIO water...

lian-li.com

I mean, a 5090 into this space? I would be quite worried about the heat by itself.

ASUS's product page is mislabeled. It is not ATX 3.1. It is 3.0, because for the PSU to be 3.1 it must have the revised 12V-2x6 socket.

Differences Between ATX 3.0 and ATX 3.1 Standards - Seasonic Knowledge Base

As PC hardware performance continues to improve, the power consumption of graphics cards and CPUs is rising, especially when high-end graphics cards are under h

knowledge.seasonic.com

 

[Scattergrunt]

Fortunately something like that already exists, but it’s unfortunate that nVidia couldn’t use it on the FE because… cost?

This is from some AIB variant:
View attachment 384256

View attachment 384257

Per pin current monitoring!

When you have this available it’s not difficult to write an instruction to shut down the load after it detects a discrepancy (beyond a threshold) in current between the pins for a couple of seconds and prompt the user to check or replace the connection.
Problem solved!

Lets not forget this is the same NVIDIA that disabled the hotspot sensor. They probably did both of these on purpose, presumably

 

[user556]

You know the difference between 12VHPWR and 12V-2x6 lies not on the cable and its female connectors, but on the GPU and PSU side with their male connectors, right?

Unfortunately for these users, that seems to be a myth.

 

[londiste]

Unfortunately for these users, that seems to be a myth.

What exactly makes you say that?

 

[jnv11]

Der8auer got the card & cable & PSU.
Observed and compared the results with his own 5090.
Finds out there are 2 wires carrying 20A in stress testing. (Other wires carried ~5A)
It makes those 2 wires extremely hot and dangerous.

Maybe TechPowerUp should investigate the 12V-2x6 connector on the RTX 5090 cards including the FE and the AIB cards, and revoke the awards from those that it can find are unsafe or have other problems with the 12V-2x6 socket or power draw. Or better yet, replace those awards with awards of dishonor or failure whenever an unsafe product is detected and proven unsafe.

As what these findings have shown me, I have ruled out buying an RTX 5090 Founders Edition for my next build and will look only at AIB cards if I want an RTX 5090.

Buildzoid made a follow-up video building on der8auer's video below:

 

[Darmok N Jalad]

As what these findings have shown me, I have ruled out buying an RTX 5090 Founders Edition for my next build and will look only at AIB cards if I want an RTX 5090.

I wouldn’t buy the FE anyway. NVIDIA goofed when they made that a 2 slot cooler. I guess they wanted to show it could be done, but it’s hotter and louder than the AIBs. It’s an innovative design, sure, but not for a 600W card.

 

[user556]

What exactly makes you say that?

The results of this long running experiment. Every time, it's the 12VHPWR plug (on the cable) that is the centre of the failings.

 

[Lordken]

same story again, and not just melting connector but how nv fanboys jumped same old song of "user error" "shitty cable"

This sadly shows that NV can sell whatever piece of junk, but as long as Jensen says "its way to be played" masses will blindly buy for whatever price.

What i find appalling watching Buildzoid video, that no one really seen this insanity from electrical perspective to point it out earlier??
Perhaps NV went with usual corporate "logic" and just fired old veteran engineers (because they are expense, workers are just evil that cost corporate money...) that build earlier power delivery circuits and replaced them with cheap graduates that didnt actually knew what they are doing (hey but they are half the price, yay!) plus they were pushed for savings, because apparently nv is such poor company that cannot afford to pay few cents on parts for cards that are overpriced and expensive...

Or maybe they just used same fake frames mentality they push with rendering, lets use just one wire and imagine/fake there are 6 Power DLSS 6

 

[user556]

It would be unlikely that the pins on the card are electrically separate. The 12V pins will be all one conductor, and the 0V pins will be all one conductor. Likewise, the power supply also has a common supply conductor at its end. That means the cards can't draw power down the cable unevenly.

Which all adds up to the cable, the plug in particular, being the only problem.

 

[Scattergrunt]

I wouldn’t buy the FE anyway. NVIDIA goofed when they made that a 2 slot cooler. I guess they wanted to show it could be done, but it’s hotter and louder than the AIBs. It’s an innovative design, sure, but not for a 600W card.

I appreciate the fact a 2 slot cooler version exists and that it can work in small builds; but just because it can doesn't mean it always should. Before CES I assumed that the SFF stuff was gonna be a separate thing that AIB's were gonna do; and it turns out NVIDIA is doing them for FE's I guess. Theres not a single non FE SFF card for the 5090 (I dont think AIB's even wanna try, and I dont blame them); and only 10 (unique cooler) SFF 5080's apparently,c ompared to the amount of 5070's that comply with SFF & the gargantuan amount of 4070/S and 4070Ti/S's that comply with it, thats pretty small.

Honestly it was kind of inevitable that NVIDIA was going to go small for the FE's where possible, considering the prototype 40 series card GN acquired that was supposedly (and probably) a prototype for the 4090 was even thicker and bigger than the 4090 we actually got, it seems like NVIDIA was always trying to go smaller and smaller even before the 50 series.

 

[Lordken]

It would be unlikely that the pins on the card are electrically separate. The 12V pins will be all one conductor, and the 0V pins will be all one conductor. Likewise, the power supply also has a common supply conductor at its end. That means the cards can't draw power down the cable unevenly.

Which all adds up to the cable, the plug in particular, being the only problem.

denial? did you watch the video? he did he measure individual cables? It directly contradicts to what you say "cards cant"

How can be copper wire faulty? it either is conduit or isnt (if it would be under spec or too thin etc, it would burn), or you think that cable manufacturer(s) are that much incompetent that they somehow manage mix wires with significantly different specs, with such difference in resistance, that one cable would allow 2amps while next wire 20 and another 8?

Also he checked his cable/connector so the plug was correctly connected , IF it would be "user error" then some other wires would show 0 amps right? which didnt, so there was contact on pins (or perhaps with lousy contact such pins would heat more, maybe like what was seen on PSU side - however in such case, with lousy contact point with higher resistance, would current rather flow via more conductive paths , other wires, so the hotspot would actually cool down as current flow switches? Sadly it wasnt possible to tell from his video if those two hot wires were same as that 150C PSU connection . But the two hot wires clearly shows they were carrying too much amps
But otherwise yes, it was said before that plug design is problem (which is still nv fault)

Or what were you trying to say? its too late, and it was kinda hard to read that sentence

 

[user556]

It's not hard for different pins to have different quality of connection. So, although the card has only one common conductor to all pins, the individual pins can be disconnected, or near disconnected.

In the demonstrated video, it's clearly two of the six pins carrying the bulk of the power. The other four are bad connections.

With the melted cable, it would have been five bad pins with only one working fully. 500 Watts is too much for one pin.

 

[AusWolf]

It's not hard for different pins to have different quality of connection. So, although the card has only one common conductor to all pins, the individual pins can be disconnected, or near disconnected.

In the demonstrated video, it's clearly two of the six pins carrying the bulk of the power. The other four are bad connections.

With the melted cable, it would have been five bad pins with only one working fully. 500 Watts is too much for one pin.

What is "near disconnected"? Electricity either flows or it doesn't.

 

[user556]

Everything has resistance, even superconductors. If the male-female isn't fitting well then you get a poor quality connection. Which shows up as higher resistance. ... Which is where the statements about "user error" comes from. The accusation is the plug is not fully seated.

However, these plugs should be able to handle not being fully seated without any undue loss of quality in the connection. And this has already been demonstrated as working fine using the 12V-2x6 plug.

 

[Crackong]

Maybe TechPowerUp should investigate the 12V-2x6 connector on the RTX 5090 cards including the FE and the AIB cards, and revoke the awards from those that it can find are unsafe or have other problems with the 12V-2x6 socket or power draw. Or better yet, replace those awards with awards of dishonor or failure whenever an unsafe product is detected and proven unsafe.

As what these findings have shown me, I have ruled out buying an RTX 5090 Founders Edition for my next build and will look only at AIB cards if I want an RTX 5090.

Buildzoid made a follow-up video building on der8auer's video below:

LOL
Nvidia made all 6 wires as one set so most of the current flow through the one wire with least resistance....

It is actually melting the one wire with the perfect contact, not the other way around.

That means the cards can't draw power down the cable unevenly.

The wires in between are separated.
And current flows through the path with least resistance.
Take the below picture as illustration
Those with ~1Ω are wires making normal contact.
3Ω is the one making bad contact
and the 0.5Ω is the one having the perfect contact

As both the card and the PSU treat the whole thing like one giant cable.
Most current flow through the path with least resistance.
So the 0.5Ω wire gets all the love and most of the current.
More current = more heat = melt.

 

[user556]

That's the plug's fault, not the card.

 

[dragontamer5788]

View attachment 384426

At 8Amps expected load per wire, 0.5 Ohms == 4V drop.

Add a few 0s there. Its maybe 0.05Ohms or less. Making miniscule differences even a bigger deal.

 

[Crackong]

Add a few 0s there. Its maybe 0.05Ohms or less. Making miniscule differences even a bigger deal.

Yes.
And this explains why der8auer got 20+Amps in 2 wires and 5 Amps in others.
Very small difference due to soldering / clamp pressure / copper impurity makes huge difference in this case

Nvidia just forgo any control and let nature do its thing on balancing between the wires.

 

[user556]

And as Der8auer also said, the cable wires are good quality. It's really only the plug that can be the problem.

 

[dragontamer5788]

Yes.
And this explains why der8auer got 20+Amps in 2 wires and 5 Amps in others.
Very small difference due to soldering / clamp pressure / copper impurity makes huge difference in this case

Nvidia just forgo any control and let nature do its thing on balancing between the wires.

Well the contact pins of Molex are usually pretty good. Copper wire is also quite consistently manufactured these days. There's plenty of charts of AWG. Finally: there are microohm meters that can test the consistency of wires and contact-pins after manufacturing. They aren't cheap, but there's plenty of ways to ensure reliable manufacturing.

In particular, the molex pins are springs that apply consistent force upon the pin. Its not the plastic housing that makes it springy, its the pin itself that's the spring and grabbing onto the contact.

 

[Crackong]

They aren't cheap, but there's plenty of ways to ensure reliable manufacturing.

To be honest.
Since both the PSU and the 5090 both treated the thing as one big wire.
The most "reliable manufacturing" thing should be...exactly one big wire.

One 6 AWG wire with good insultation should do the job just fine.

 

[dragontamer5788]

To be honest.
Since both the PSU and the 5090 both treated the thing as one big wire.
The most "reliable manufacturing" thing should be...exactly one big wire.

One 6 AWG wire with good insultation should do the job just fine.

You need a 2nd wire for ground.

Also, a 6AWG wire is ~5mm in diameter. Your typical wall wires in a house are 12AWG (2mm) in diameter. 6AWG is much bulkier, expensive. In fact, I've placed two circles next to this 12vhpwr drawing to give you an idea of just how big 6AWG wire is.

That's just the wire, not including the pins or connector needed to work with such a wire. Any electrician who has worked with this knows how hard it is to work with. Its not easy bending or moving copper cables that thick around.

 

[Crackong]

Also, a 6AWG wire is ~5mm in diameter. Your typical wall wires in a house are 12AWG (2mm) in diameter.

They uses 110V/220V, that's different.

5090 runs on 12V 50A, and the working temperature >60C.
So 1x 6AWG, or 2x 10AWG
Give it some safe room and the solution could be just scale up the original 6 pin design with 10 AWG wires and connectors

 

[user556]

If the plug works correctly then 12 pins is fine. It's been fine for the 6-pin and 8-pin GPU plugs. 8 pins for the CPU power plugs are doing just fine too.