• 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

kalston said:
I'm a little confused. PCI-E 8 pins is 150w, so when pushing 900w they were out of spec for all wires? (nearly 300w per wire assuming perfect balancing?!)

Did the 8pin connectors have that much margin or did they maybe use a CPU power connector instead? Those EPS ones are rated much higher (over 300w) and I suppose they could be modded to work on a GPU.

edit: on their Youtube channel I see custom connectors, not PCIE 8pin connectors. Still neat, don't get me wrong. But seems like some things got lost in translation maybe.
Click to expand...
It looks like a lot more margin on the PCI-E 8 pin connectors. The test is what temperature they are when under load. 35C sounds really good as the concern here is mostly resistive heating as I don't think 35C is high enough to affect metal platings. Keeping it to 25C is even more better.
 
it’s going under the knife right now!!!

unknown.jpeg
IMG_6804.jpeg
IMG_6803.jpeg
 
kalston said:
I'm a little confused. PCI-E 8 pins is 150w, so when pushing 900w they were out of spec for all wires? (nearly 300w per wire assuming perfect balancing?!)

Did the 8pin connectors have that much margin or did they maybe use a CPU power connector instead? Those EPS ones are rated much higher (over 300w) and I suppose they could be modded to work on a GPU.

edit: on their Youtube channel I see custom connectors, not PCIE 8pin connectors. Still neat, don't get me wrong. But seems like some things got lost in translation maybe.
Click to expand...

The PCI-E 8 pins were designed to carry over 300W each but were limited to 150W. They were designed that way for safety and so you could daisy chain 2 connectors on the same card.
 
Starfalcon said:
The PCI-E 8 pins were designed to carry over 300W each but were limited to 150W. They were designed that way for safety and so you could daisy chain 2 connectors on the same card.
Click to expand...
Yep it had a very nice safety factor you could cut a wire and the others would still not burn....the new one if you run full power it will degrade and burn
 
TXE36 said:
While the mechanics of this failure is simple, the cause is likely complex. The bottom line is this connector and/or implementation doesn't work and may be unsafe. This problem cannot be fixed by hanging some do-dad on the power connector.

The mechanics is too much resistance in the connections at the pin and likely also the wire crimps for the amount of current the pin is passing. In fact, that's the only thing that can be happening that is melting the wire insulation and plastic connector housings. It is extremely difficult to protect against this because it is likely the amount of current is in spec for the video card. Things like fuses and circuit breakers are designed for fault conditions - this isn't a fault condition. If the video card broke up the phases properly, it likely could detect the voltage drop of a overly restive pin and shut down the card. However, splitting the phases costs more money and provides fewer leather jackets. (The more I'm thinking about this, the real cause does appear to be lack of power phases on the 4090 and 5090s, see below).

The cause is a whole other kettle of fish. Based on reports of connections lasting a long time and then smoking, I'm suspicious of the wires themselves and the plating on the connector sockets and pins. This plating is designed to maintain low resistance over a specified number of mating cycles and atmospheric exposure. I hope they are using copper wires and not some garbage like copper coated aluminum. It could also be simple overload of the per pin current rating (see below). It can also be combinations of all of these sorts of things, hence complex.

A common way to enshitify a connector is to go cheap on that plating. Another way is making the mechanical tolerances more sloppy. The causes I'm mentioning here are speculation, but I'm sure the connector manufacturer knows what is going on and there have been uncomfortable meetings about this.

IMHO, Nvidia has a very big problem here with a connector that now has a large installed base. I can see a big recall in their future and heaven help them if this manages to burn a house down.

I watched the first few minutes of the Buildzoid video which was inspired by a der8auer video and it does look like the fault here is Nvidia is using the connector wrong and not balancing the load. der8auer's IR video clearly shows only two wires/pins are providing the bulk of the power to the 5090 -- this is really, really poor engineering:

View attachment 825240

I skimmed the Buildzoid video and he is right on target. Remember on previous Nvidia cards like the 1080TI, if you forgot to plug in a power connector, the card would not go into 3D mode. The card could sense this because of the split power phases in the voltage regulator.

Providing 600W of power at 12V requires 50A of current, so one has to spread this out over several wires. The problem is, the current magnitude is high enough here that it gets sensitive to the resistance of each wired connection and prefers the ones with lower resistances, this can over load them and cause their resistance to increase, making the problem worse. Split power phases forces the current to split rather than relying on the resistances of each connection to be the same.

Going to a 24V feed instead of 12 would help too as only 25A would be required at 24V for a 600W load. This kind of change has been done before - the development of the ATX12 connector for the CPU. This was done because feeding the CPU regulator with 5V was burning up motherboard ATX power connectors. ATX12 was easy as it's just another connector. 24V is tricky because the voltage rail itself is new.

Multi-kilobuck video cards designed with power connector/feeds like this is a really sad state of affairs. Enshittification is everywhere these days.
Click to expand...

"Maybe Buildzoid ALSO knows nothing about how computers work!".

- Nvidia apologists, probably
 
AVATARAT said:
Be careful when adding a second connector to the Gigabyte 5090, especially if you plan to use it with a power consumption above 600 W:

View attachment 826279

Source
Click to expand...
As an experienced engineer without any other data to go by, feeding all those MOSFETs from that one tiny connector on the upper right just looks wrong.

Not commenting on the circled crater, just using the picture because it is convenient. However, that blowed up real good!
 
If it's the card I'm thinking about je was trying to push 1kw though it of course something on the vrm is going to pop
 
The video is old; you can watch it here, as I linked to it in my previous post.

I know people who actually use this mod with a second connector—generally speaking, this is risky, but it distributes the power between the two connectors, which makes it harder for them to melt.

It’s only possible on the Gigabyte model because there are already wired paths on the board, and the 12VHPWR port is simply soldered to it.
 
It's too bad it does not disconnect the 12V when the temp gets too high.

One positive thing is it looks like the burned pins, if they happen, will happen to the power supply cable and this gizmo rather than pricey video card. The resistance seen by the video card connected should be pretty balanced with this thing in place.
 
serpretetsky said:
Why? The pins on this device could have just as much resistance mismatch as the cable.
Click to expand...
See this is what I don't think people understand about an electrical circuit.
More connections = worse by increase in resistance = more heat = higher likelihood of a connector failing but now the kicker instead of a two x of one or both ends fails we now have a x4 chance that any one or more can be out of spec making more heat and burn.

It's physically impossible to make an electrical circuit more reliable by adding more connections from one power source that's why I think if you are worried about it the only thing that makes sense is a PSU that has actual per pin monitoring not adding something with the hope it will trip or melt before the card melts.
 
^^^ correct, it would have been better for Nvidia to solder the wire whip directly to the card with no connector, thus connecting to the PSU would be the only connector. eliminate one connector and there wouldn't be a worry about bending the cable/connector at the video card, because once it is soldered to the card it wouldn't matter.
 
tunatime said:
More connections = worse by increase in resistance = more heat = higher likelihood of a connector failing ...

It's physically impossible to make an electrical circuit more reliable by adding more connections from one power source ...
Click to expand...
I agree if you mean in series. I Disagree if you mean in parallel.
 
serpretetsky said:
Why? The pins on this device could have just as much resistance mismatch as the cable.
Click to expand...
They might, but I think it is less likely. Pins soldered to a fat trace on a circuit board are going to be more uniform in resistance than relatively long wires crimped on to pins. You are correct in that if the resistance difference is from the contact on the pins themselves, this won't make a difference.

The whole design is sub-optimal, and as I said before, it cannot be fixed by adding some do-dad on the connector.

About the only thing that will help is the ugly soldered pigtails running to an appropriate number of PCIE 8 pin connectors. Other approaches will require a board spin. Even with the soldered connectors, the boards still don't have the multi-phase power feeds. They could burn up PCIE 8 connectors too.

Nvidia screwed the pooch on these things big time. Wouldn't surprise me if several people have been fired over this.
 
In the end, the most practical thing to do here is just turn the power down by dialing back the cards a bit.

I don't have any experience beyond Pascal, but on Pascal, the stock Nvidia bios really pounds the shit out of the card to get that last bit of performance. One can dial back 25-30 watts of power on a Titan Xp with no loss of performance by undervolting and overclocking. If one is willing to give up 2-3% of performance, one can save 50 watts, which is 20% of the original TDP of 250W.

Nvidia has the advantage that they can tweak the bios itself and this is several orders of magnitude easier and cheaper to do than recalling and replacing these cards. If Nvidia were a normal company, this sort of thing could bankrupt them.

Is there a way to dial the cards back in Afterburner?
 
  • Like
Reactions: Niner
like this
TXE36 said:
In the end, the most practical thing to do here is just turn the power down by dialing back the cards a bit.
Click to expand...
Sure. But who buys a 90-series card to be practical? I think we've already had at least one person on this site say he didn't want to do that.
 
TXE36 said:
In the end, the most practical thing to do here is just turn the power down by dialing back the cards a bit.

I don't have any experience beyond Pascal, but on Pascal, the stock Nvidia bios really pounds the shit out of the card to get that last bit of performance. One can dial back 25-30 watts of power on a Titan Xp with no loss of performance by undervolting and overclocking. If one is willing to give up 2-3% of performance, one can save 50 watts, which is 20% of the original TDP of 250W.

Nvidia has the advantage that they can tweak the bios itself and this is several orders of magnitude easier and cheaper to do than recalling and replacing these cards. If Nvidia were a normal company, this sort of thing could bankrupt them.

Is there a way to dial the cards back in Afterburner?
Click to expand...
1_rick said:
Sure. But who buys a 90-series card to be practical? I think we've already had at least one person on this site say he didn't want to do that.
Click to expand...
“another media outlet/reviewer had their 16pin melted. this time, it was expreview 超能网, a well-known tech media/reviewer in china. funny enough, they spotted it while analyzing "dlss5 melting 16pin". zotac 5090dv2 solid ocseasonic vertex px1200 limited editionnvidia power capture analysis tool (pcat) used16pin melted (pcat end)超能网 expreview's statement in the video:""Because the power supply and PCAT kit cables have been in use for a long time, the overheating and melting issue with the cables was only discovered during the recording of this video. It is impossible to determine whether the problem was caused by cable aging or prolonged full load. This video segment is only intended as a reminder; please pay close attention to the 16-pin power connector and cables when using your graphics card at high power for extended periods.""“

https://x.com/unikoshardware/status/2101030168676667559
 
1_rick said:
Sure. But who buys a 90-series card to be practical? I think we've already had at least one person on this site say he didn't want to do that.
Click to expand...
Shrug: Sounds better to me than trashing a several thousand dollar video card.
 
  • Like
Reactions: erek
like this
1_rick said:
I'm not gonna disagree, but I'm also not in the market for "gotta have the top of the line no matter the cost" video card.
Click to expand...
Ha! Neither am I. I have no interest in kilobuck video cards.

I've watched this thread grow for some time in passing because I don't have a dog in this hunt beyond being interested in the topic and I have some experience with switching power design. What finally got me to comment was the denial going on and the fact there is a market for hundred dollar do-dads to band aid what is obviously a piss poor design. Throw in a helping of "mines fine, so you must being doing something wrong" on top of it.

What finalized my opinion was the removal of the power feed phases on the video card it self - a true shake your head moment. So Nvidia's plan was to shrink the power connector and to ensure success, also remove the power feed phases? With the amount of margin these cards bring in, it is an astounding amount of risk for pennies in return.
 
You must log in or register to reply here.
Back
Top