It's happening again, melting 12v high pwr connectors

[wNotyarD]

The connection becomes DRASTICALLY worse after replugging just a couple of times, due to the connection relying on tight pressure to make the connection on the tiiiiny connection surface.

Yes, but that also only becomes so much of an issue because the power draw is much higher than it should be rated for. Limit each cable/connector to 300~350W and it'd become much lesser of an issue.

 

[Dragam1337]

Yes, but that also only becomes so much of an issue because the power draw is much higher than it should be rated for. Limit each cable/connector to 350W and it'd become much lesser of an issue.

Yes and no - you'd still end up in situations like derbuer, where several connectors aren't making a proper connection due to the super weak connector not being able to handle being replugged (unlike pcie 8), leading to 1 or 2 wires to carry all the amps - and while that would be less problematic at 350w, it could still lead to issues. We have seen 4080's with melted connectors after all, even if not nearly as many as 4090's.

 

[GamreCaleb]

For me, I find great pride in plugging in 3, 8 pin connectors into my GPU. So why do we need to make smaller plugs? Or is it just me?

 

[Bomby569]

For me, I find great pride in plugging in 3, 8 pin connectors into my GPU. So why do we need to make smaller plugs? Or is it just me?

nvidia elitists can't plug more than one cable

 

[wNotyarD]

For me, I find great pride in plugging in 3, 8 pin connectors into my GPU. So why do we need to make smaller plugs? Or is it just me?

I have no issues using 3 PEG connectors, but 4 or more would become overwhelming, not to mention PSUs need to have that many connectors as well.

 

[ShrimpBrime]

I have no issues using 3 PEG connectors, but 4 or more would become overwhelming, not to mention PSUs need to have that many connectors as well.

My HX psu built a decade ago comes equipped with 4 8x pin connectors. This should be no problem for modern power supplies 1kw and higher. Imo.

 

[JustBenching]

Since a 5090 can pull up to ~700w sustained, it would need 5x8pins to be within spec. Isn't it just both better and safer to have 2 or 3x 12vh instead? So why is everyone complaining about the damn cable, I don't get it.

 

[Bobaganoosh]

Everything is relative. For it's intended purpose, it's a really crappy connector design. If it was designed for 300W, it'd be quite robust and the bad tolerances we see wouldn't cause an issue 99.99% of the time. Because there's no headroom, any tolerance issues or malformations can cause a problem.

These stamped sockets are the cheapest of the cheap. The molded connectors are horrible to maintain quality on. They are complex little shapes and the tooling wears out over time, so you end up with part to part variation in the molds in general, and then over time you get additional slop as the tooling wears. Then the sockets that have their own tolerances due to the nature of bent sheet metal because these are just stamped contacts.

There's a solution that's been readily available in connectors for decades: machined pins and sockets. They're actually so readily available at this point that they're not even that much more expensive, especially when we're talking about GPUs over $1k. I just don't understand why they went with such a cheap crappy design when they knew they were going to push right up against this power limit and they knew the risks involved. It's so bad that my only assumption is that management made the engineers leave the room after they complained too much. I'm a biased engineer though, so of course that's my opinion lol.

Edit: just to clarify something: I keep seeing comments in here about Brand X or Brand Y having an issue or "well mine is perfect"....this is a design problem as in if everything is manufactured flawlessly, the cables are ok. The design does not properly accommodate for any manufacturing tolerances or part-to-part differences. Any misalignment can cause problems. It's easy to say "oh well that's just a bad part", but if that part meets the spec, it's really a bad spec and a bad design for the use-case. We're bound to see good and bad parts from any brand trying to make parts at this point because the design itself and the means to manufacture it will lead to variations and issues eventually.

 

[LittleBro]

No, using multiple bad connectors is not the way. 12VHPWR/12V2x6 was specified/designed to deliver rated 600W of power. The thing is that the design sucks. Design will still be bad even if you use more bad connectors. Does not change a fact. Then it absolutely fails its purpose to combine multiple cables to just one.

Also, not every PSU has more than one 12V2x6 connector. What are you gonna do when your GPU requires 2 such connectors? Use Y-splitter? It's designed to be one connector and it would be okay had they not shrinked it too much. A bit bigger, more robust 12V6x2 would do the trick.

Using argument to multiply bad connector to help on this matter is as good as supporting a roof terrace with many glass pillars instead of few properly sized metal or concrete ones.

Those who think they are fine when they don't see any melting yet should read this:

Official RTX 4090 power cable found melted by reviewer 2 years later — card functioned fine despite hidden melted connector

Nvidia is kicking it old-school with Molex-to-SATA levels of fire hazard.

www.tomshardware.com

Now don't you ever try to unplug the bloody connector just to be sure! Let's just wait for melting to appear, so you won't waste a plugging cycle this way! FFS.

 

[OneMoar]

the connector ISN`T the problem yall are stuck in a loop talking about the connector the issue isn't the connector is the FE cards having crap power delivery drawing all of there power of 2 wires because there is no bias control stop talking about the connector its not the problem in this case

 

[AlonsoTy]

I wonder if having connected the two rows of contacts together on the GPU side would not have at least mitigate the problem...
1x6(12V) and 1x6(0V)

They aren't actually connected together... right?

the connector ISN`T the problem yall are stuck in a loop talking about the connector the issue isn't the connector is the FE cards having crap power delivery drawing all of there power of 2 wires because there is no bias control stop talking about the connector its not the problem in this case

i think so

 

[phanbuey]

the connector ISN`T the problem yall are stuck in a loop talking about the connector the issue isn't the connector is the FE cards having crap power delivery drawing all of there power of 2 wires because there is no bias control stop talking about the connector its not the problem in this case

I think that's what people mean when they say "the connector" though. You're right its the power delivery across the connector, but the change in connector is when the power delivery changed as well. So it's more "the change to this standard" anger.

The 2080TI melted its ram chips, but never melted the 8pins.

 

[Caring1]

the connector ISN`T the problem yall are stuck in a loop talking about the connector the issue isn't the connector is the FE cards having crap power delivery drawing all of there power of 2 wires because there is no bias control stop talking about the connector its not the problem in this case

Nope, that's not it.
The power delivery is unbalanced due to the CONNECTOR.

 

[MxPhenom 216]

the connector ISN`T the problem yall are stuck in a loop talking about the connector the issue isn't the connector is the FE cards having crap power delivery drawing all of there power of 2 wires because there is no bias control stop talking about the connector its not the problem in this case

I was just about to say. The big issue here is the lack of load balance across the 12v wires.

Funny enough, Asus actually is doing gods work here with their newer PSUs and paired with their cards. When using both together they have load balancing with unique circuitry from PSU side and GPU side to balance current going through each wire, and GPU Tweak software that can monitor each individual wires load.

I don't think other PSUs and cards have this kind of thing built into them.

Nope, that's not it.
The power delivery is unbalanced due to the CONNECTOR.

What do you think the CONNECTOR is actually doing here to contribute to this?

 

[Knight47]

Funny, derbuers corsair cable is 2x6...

It's the old unsafe 12vhpwr, he just had insane luck it was working fine for that short time.

 

[wNotyarD]

What do you think the CONNECTOR is actually doing here to contribute to this?

There's two sides here. If load balancing was mandated, it'd allow for flawed/worn cable connectors here.
Since it isn't, mating must be immaculate. As it frequently isn't, there's resistance variance in a connector which has no margin for flaws at the power required from it.

 

[BoggledBeagle]

The problem is caused by a faulty cable with faulty connectors, that connect all the parallel wires differently, therefore two different things can happen:

1. Too much current runs through some wires.
2. The contact area is too small, current density too high causing too much heat generation and heat damage in the connector.

Would anybody complain, if the card malfunctioned or was damaged, when a PCI-E connector in the motherbord was damaged? No, of course not. That would be extremely dumb, right?

 

[Bobaganoosh]

the connector ISN`T the problem yall are stuck in a loop talking about the connector the issue isn't the connector is the FE cards having crap power delivery drawing all of there power of 2 wires because there is no bias control stop talking about the connector its not the problem in this case

It's a combo. If the connector was reliable, they wouldn't need any of that extra circuitry. They tried to save money on both ends and it's costing them. The load balancing across wires shouldn't be necessary at all if you could rely on the cable, but we have an unreliable connector design. One way to catch a bad connector is with load-balancing or at least line monitoring circuitry that throws an alarm. It'll just tell you that your connector/cable is bad...which is helpful, but doesn't fix the problem that there will be more bad connectors/cables with this inappropriate design. So your argument boils down to "it's not the connector, we need better circuitry to compensate for the bad connector"...which contradicts itself.

Edit: If you really wanted redundancy, and sometimes you do reach out to the department of redundancy department for these things, you go with the "belt and suspenders" design mentality and you design a quality connector-scheme in addition to circuitry that detects issues properly as well. Nvidia did neither here...some AIBs are doing one because they have no power over the other...but it's still bad.

 

[ir_cow]

Seems to me in summary of der8auer newest video is if a pin or wire is damage either at the PSU side or in the GPU side, the load will become unbalanced leading to over 9a and eventually cable melting.

 

[SmookinJoe]

continued....

Hmmm
I wonder if this is one of the arguments that came up with Nvidea and EVGA when they said they weren't getting respect...
I am referring to the load balancing capable built into the 3090 but is non existent in the 4090+ that the video mentions...

End result being EVGA is not associated with this 12 pin BS....

 

[TheDeeGee]

Maybe we should also have listened to Intel 2 years ago

Intel Recommends 4-Spring 12VHPWR Connector Design Over Less Reliable 3-Dimple Designs - Pokde.Net

But perhaps it's also good we didn't because otherwise this criminal negligence of NVIDIA wouldn't have surfaced.

 

[ty_ger]

If you try pushing 700w out of an 8 pin cable it will fail, no question about it.

But we wouldn't right? Because they are properly rated to have a huge safety margin and the manufacturer would use multiple to have a product that isn't way out of certified limits and thus a liability.
The problem here is that the 12-pin connector is not properly rated and NVIDIA refuses to acknowledge that and adjust as needed. From NVIDIA's perspective, it is someone elses problem that their sexy design fails when any minor abnormality occurs with the connections.

 

[evernessince]

What "protections" do you want? Individual heat sensors on every cable pin? Can you imagine the cost of that?

Did you read the thread?

How about current balancing capabilities like what 3000 series and older cards have as a start and a connector with decent safety margins as a second.

I don't get why so many people jump into this thread without having followed the discussion at all. The things want from these cables aren't radical, they are common sense things we've had for decades.

We haven't seen any 12v-2x6 fail, neither damaged or undamaged ones. All failed cables are old 12vhpwr ones.

You clearly haven't been looking at all the reports but to be fair there have been a ton of them lately.

Seems to me in summary of der8auer newest video is if a pin or wire is damage either at the PSU side or in the GPU side, the load will become unbalanced leading to over 9a and eventually cable melting.

That aligns with what buildzoid pointed out. Nothing has a 0% defect rate and thus problem with this cable has the potential to cause catastrophic failure.

I do also believe there's another mode of failure though, given the number of people finding their connectors partially melted after a few years of service.

 

[SmookinJoe]

Another point..
Guys are saying they should check their connectors on their running 4090's.
How by removing them?
DON'T DO IT! (IMHO)
If it is running fine leave it alone...
Check with a heat sensing device.
Check that it is seated, not warm to touch...anything that doesn't involve wiggling and unplugging and replugging...
Gamers Nexus showed us the design has a very low lifespan when it comes to connects and disconnects of 12pin connectors..
UNLESS YOU SEE AN ISSUE..and need to further investigate..
Don't fix what isn't broken..
If I am concerned, I have a cheap little lazer thermometer and I probably NEED to check it more often.
That is my bad...
Just thinking out loud guys...
OK...Just understand I am trying to REMIND you what we already know.
Just needed to voice my opinion to help prevent issues you might cause.

 

[phanbuey]

Another point..
Guys are saying they should check their connectors on their running 4090's.
How by removing them?
DON'T DO IT! (IMHO)
If it is running fine leave it alone...
Check with a heat sensing device.
Check that it is seated, not warm to touch...anything that doesn't involve wiggling and unplugging and replugging...
Gamers Nexus showed us the design has a very low lifespan when it comes to connects and disconnects of 12pin connectors..
UNLESS YOU SEE AN ISSUE..and need to further investigate..
Don't fix what isn't broken..
If I am concerned, I have a cheap little lazer thermometer and I probably NEED to check it more often.
That is my bad...
Just thinking out loud guys...
OK...Just understand I am trying to REMIND you what we already know.
Just needed to voice my opinion to help prevent issues you might cause.

Mine ran fine for 1.5 years then melted...