NV 12VHPWR issues revisited

[BFG10K]

He gets a 4090 almost every day with a melted connector. He also says it's common for the connector to fully snap off 4xxx cards.

These videos are only a few days old. Why aren't tech sites reporting these ongoing incidents? What the heck is going on here?

This abomination of a connector is clearly defective design, yet the whole fiasco seems to have been covered up.

 

[jpiniero]

Cuz it's Youtube. Probally fake nooz.

 

[BFG10K]

Cuz it's Youtube. Probally fake nooz.

LOL. You didn't even watch the videos before posting, did ya?

The guy's soldering and PCB skills are unreal. Nothing fake about him.

 

[KompuKare]

He gets a 4090 almost every day with a melted connector. He also says it's common for the connector to fully snap off 4xxx cards.

These videos are only a few days old. Why aren't tech sites reporting these ongoing incidents? What the heck is going on here?

This abomination of a connector is clearly defective design, yet the whole fiasco seems to have been covered up.

Media cover-up is one thing

Far more concerning are the fans of that company with the green logo going around and basically doing a Steve Jobs excuse of "you're holding it wrong".

Still so many posts everywhere putting it all down to use error.

 

[WelshBloke]

300 watts and high transient spikes is a lot for any connection!

I've always thought that connections in PCs have always been a bit shonky (I'm looking at you molex), I guess that the difference now is that we are pumping ungodly amounts of power through them.

 

[Mopetar]

At this point it's just old news, so no one really cares.

 

[Leeea]

100% user error.



The user purchased a nvidia product. They made an error.

 

[alcoholbob]

Meanwhile MLID continues to parody the official Nvidia line “it’s something like a couple dozen people tops that had a problem with the 12V connector with Lovelace. Basically not a problem”

 

[BFG10K]

300 watts and high transient spikes is a lot for any connection!

The connector is actually "rated" for 600W.

Meanwhile MLID continues to parody the official Nvidia line “it’s something like a couple dozen people tops that had a problem with the 12V connector with Lovelace. Basically not a problem”

NV says ~50 cases. So either every one of those failures is sent to NorthridgeFix to repair, or the numbers are higher.

 

[A///]

Meanwhile MLID continues to parody the official Nvidia line “it’s something like a couple dozen people tops that had a problem with the 12V connector with Lovelace. Basically not a problem”

parrot not parody

 

[Mopetar]

Calling MLID a parrot is insulting to parrots.

 

[Aapje]

We've seen that it can require a lot of force to seat the connector, so customers now have to basically press hard enough to fully seat the connector, but not so hard that they snap it off. Either not seating it fully or breaking the connector can be called user error, but the question is whether it's reasonable to expect people to be able to do this.

 

[cmdrdredd]

At this point it's just old news, so no one really cares.

Plus there are cables that aren't the stiff, unbendable Nvidia adapters. PSU manufacturers like Corsair offer their own cables that plug directly into your PSU and have individually sleeved wires so you can push them out of the way. Cablemod offers their own adapter that is also better than the Nvidia provided cables. I believe they even have a 90 degree version that doesn't stick out from the card at all. The whole problem is the connector being pulled on and stressed by a bad cable design in a case it just doesn't fit in without pushing that cable a bit and a stiff connector. Aftermarket options are just better and don't have these problems.

There are still people who will bring it up at every opportunity in order to crap on Nvidia though. How are those burning out 7900s with faulty cooling systems from AMD going? Same thing if we brought that up at every opportunity.

 

[cmdrdredd]

We've seen that it can require a lot of force to seat the connector, so customers now have to basically press hard enough to fully seat the connector, but not so hard that they snap it off. Either not seating it fully or breaking the connector can be called user error, but the question is whether it's reasonable to expect people to be able to do this.

As an owner of a 4080 and using Corsair's premium 12VHPWM adapter, the cable is super easy to fully seat in the card. It's a bad adapter design from Nvidia in the box of the GPU.

 

[Aapje]

The whole problem is the connector being pulled on and stressed by a bad cable design in a case it just doesn't fit in without pushing that cable a bit.

That's not the whole problem. We've seen that the fit can be so tight that it requires extreme force to fully seat the connector, which has resulted in people plugging it in only partly.

Connectors have melted in situations where there was no pressure on the cable.

As an owner of a 4080 and using Corsair's premium 12VHPWM adapter, the cable is super easy to fully seat in the card. It's a bad adapter design from Nvidia in the box of the GPU.

When you have a connector and a socket to fit it in, a proper fit requires the two to be sized correctly in relation to the other. You can't just declare that the connector is the issue, when the socket can also be too tight.

Of course, it is far easier to replace a cable than to replace a socket.

Ultimately, the question that has still not been answered is whether the standard has correct sizes (it should have both an intended size and allowed deviation, since no production process is perfect); and whether the connectors and sockets follow the standard.

 

[H T C]

The whole problem is the connector being pulled on and stressed by a bad cable design in a case it just doesn't fit in without pushing that cable a bit and a stiff connector.

As i understood it, the problem was that the design allowed for the cable to LOOK LIKE it was fully plugged in with the sensor signaling that IT WAS connected, when in fact IT WAS NOT properly connected. This resulted in a build up of heat that would eventually lead to the cable failures.

IIRC, one of Gamers Nexus video demonstrated this, but i don't recall which one exactly: MIGHT BE the one where they used that external lab for the 1st time, but i'm not sure.

Ofc, the placement of the plug on the card contributes A LOT for the issue, as it's not easily visible when the cable isn't properly connected.

 

[Leeea]

It's a bad adapter design from Nvidia in the box of the GPU.

That is part of it.

It is also the sense pins signalling it is fully plugged in when it is not close to clicked in.

The plug is positioned on the official cards such the user cannot really see what is going on easily.


It is a couple issues together that are going to come together and make it a ghost that haunts this gen of cards.

 

[BFG10K]

A few days later, he's had another four 4090s with melted connectors. One arrived at the end of this video and it had melted with a fully-seated 90 degree adapter.

This isn't user error. This is a defective product.

 

[Aapje]

A few days later, he's had another four 4090s with melted connectors. One arrived at the end of this video and it had melted with a fully-seated 90 degree adapter.

This isn't user error. This is a defective product.

And what is extra worrying that most 4090's should still be under warranty, so logically, this is the tip of the iceberg. Only people who did a water cooling mod or something else to void their warranty should commission a repair themselves. The rest should be RMA'd.

But what happens when these cards run out of warranty and the same issues keep happening? Then you have to pay quite a bit of money to get a repair done, even though this is a defective product that should be recalled.

 

[BFG10K]

He worked on the card with the 90 degree Cablemod adapter. Despite firmly plugged into the GPU side, melting fused the adapter to the GPU.

But there was no damage to the adapter -> cable connection. This proves the problem is the connector on the GPU side, plain and simple.

And yet another burnt 4090 showed up while he was working. "User error", gg.

 

[Aapje]

The user error is buying a card with a 12VHPWR connector.

 

[IEC]

Oof, the fact that the 90 degree Cablemod connector was used and yet the card's built-in connector appears to be at fault really does suggest an inherent problem with (at least some of) the 12VHPWR connector designs. Might be why Intel recommended the 4 spring design over the 3 dimple?



As a side note his solder looks really clean.

Original:


NorthridgeFix, postfix:

 

[Aapje]

It turns out that the cards all came from Cablemod and were sent to this one repair guy for repair. They were also saved up for a while, so it doesn't seem that the problem is as big as the repair guy thought.

 

[DeathReborn]

It turns out that the cards all came from Cablemod and were sent to this one repair guy for repair. They were also saved up for a while, so it doesn't seem that the problem is as big as the repair guy thought.

Also some were genuine user error but still Cablemod bought them from the end user effectively & paid for the shipping. The fixed cards get given away, so all in all a storm in a teacup.

 

[Mopetar]

GN Steve talked about it again recently.