Nvidia 40-Series power connectors, melting, and burning.

[Quackers101]

17 minutes ago, ShawtyT30beTHICCC said:

Apparently these are melting because people aren't plugging them in all the way. There was an interesting article from Johnny Guru on the subject. You really have to push these 16 pin connectors in HARD

not quite, the cable/connector could get loose to, that even if connected "correctly" it's incorrectly attached.

[Kisai]

13 minutes ago, Quackers101 said:

not quite, the cable/connector could get loose to, that even if connected "correctly" it's incorrectly attached.

Thermal expansion is a thing.

[Uttamattamakin]

On 11/4/2022 at 12:36 PM, jos said:

It is not just power adapter even native connector melts. So issue runs more deeper

https://www.tomshardware.com/news/rtx-4090-native-16-pin-melting

 

This is where the fundamental physics of the situation I mentioned early on come in and asserts itself. 

Pretend we weren't talking about electricity for a moment.  Suppose these were rockets.  They take power, channel it through a narrow opening and out to do useful work.  Rocket goes up. 

IF your rocket is a small scale model that can be done with balsa wood, and cardboard by a child.  
If your rocket is SLS or Falcon 9  then it has to be PRECISELY DONE or else BOOOOOOOOOOMMMMM.  

 

The 4090 and ATX 3.0 12VHPWR or whatever we want to call it is at a point where everything has to be done VERY precisely and very carefully and perhaps overengineered so that it is fool proof.  Just having the cards have 4 of the old 8 pins on them would've been better.  We'll see.  We'll see with RTX 6000 ADA generation has.  I'd not be surprised if it is using the old connectors, with high quality 2 slot blower coolers.  

[leadeater]

29 minutes ago, Uttamattamakin said:

We'll see with RTX 6000 ADA generation has.  I'd not be surprised if it is using the old connectors, with high quality 2 slot blower coolers.  

They won't, Nvidia has fully moved away from PCIe 8 Pin last generation, they aren't going back. It's also 300W TGP anyway. I think they used EPS 8 Pin.

 

Quote

The board is rated at 300W, so you’ll need a fairly hefty PSU in your workstation. It gets its power from a special 8-pin connector, which is connected to two standard 8-pin PSU connectors via an adapter.

 

More likely they will go with the new 12 Pin if they make a change.

 

But unless their site is not accurate then it's using the same EPS 8 Pin power connector

 

EPS 8 Pin is 4 conducting 12V pairs and PCIe is 3 conducting 12V pairs, that's why it can carry more power.

 

[HumdrumPenguin]

I think every time a card is sent for RMA because of burnt connectors, the AIB or NVIDIA will replace it for the updated version that allows for an easier connection with the cable. 

[Kisai]

1 hour ago, HumdrumPenguin said:

I think every time a card is sent for RMA because of burnt connectors, the AIB or NVIDIA will replace it for the updated version that allows for an easier connection with the cable. 

So they changed the connector so the 4-pin is shroud is closer to the edge of the 12pin? What does this solve?

 

https://www.extremetech.com/gaming/340772-corsairs-psu-chief-weighs-in-on-melting-rtx-4090-adapters

Quote

 According to his data, the situation might be a simple matter of what folks in the IT world call “ID10T.” That is, he thinks connectors are probably melting because they weren’t plugged in all the way.

...

 He notes this is not necessarily their fault, because the new adapter is apparently difficult to insert fully. This is due to the new sideband rails that communicate with the GPU, he says. He also notes we’re seeing more melted adapters versus ATX 3.0 cables just because the former are more prevalent.

 

I guess the new version kind of tries to solve it giving the person plugging it in more contact area, and kinda confirms multiple posters, including myself's theory that the connector really isn't deep enough to make enough contact.

 

[CarlBar]

6 hours ago, Kisai said:

So they changed the connector so the 4-pin is shroud is closer to the edge of the 12pin? What does this solve?

 

https://www.extremetech.com/gaming/340772-corsairs-psu-chief-weighs-in-on-melting-rtx-4090-adapters

 

I guess the new version kind of tries to solve it giving the person plugging it in more contact area, and kinda confirms multiple posters, including myself's theory that the connector really isn't deep enough to make enough contact.

 

 

Does indeed looks like they're trying to address a seating issue with the lower set of pins. I'm guessing those are sense pins? Not sure how it pins out as a connector TBH. Until this blew up i was barely paying attention to it.

[leadeater]

23 hours ago, CarlBar said:

Until this blew up

HAH! Nice 

[CarlBar]

1 minute ago, leadeater said:

HAH! Nice 

 

Not intentional actually. But thanks all the same.

[leadeater]

Just now, CarlBar said:

 

Not intentional actually. But thanks all the same.

That's what made me laugh so hard, because I knew it wasn't intentional. The best kind.

[Kid.Lazer]

On 11/8/2022 at 12:52 PM, leadeater said:

And for those wondering why not just use fewer much larger cables, voltage drop.

I have to go back to this statement because it's incorrect. Voltage drop is a function of amperage and the cross-sectional area of a given conductor, regardless of it being a multi- or single-conductor. A single 8AWG would lose approximately 1.05% at 50 amps load over 2 feet with it's 8.37mm² of area. 6 sets of 16AWG would also lose approximately 1.12% at 50 amps combined load (8.33 amps singularly) over 2 feet with it's 7.86mm² (1.31mm² each) of combined area.

[Dogzilla07]

 

Finally the pages for individually sleeved different colored Corsair 12+4 pin cables are up

[leadeater]

6 hours ago, Kid.Lazer said:

I have to go back to this statement because it's incorrect. Voltage drop is a function of amperage and the cross-sectional area of a given conductor, regardless of it being a multi- or single-conductor. A single 8AWG would lose approximately 1.05% at 50 amps load over 2 feet with it's 8.37mm² of area. 6 sets of 16AWG would also lose approximately 1.12% at 50 amps combined load (8.33 amps singularly) over 2 feet with it's 7.86mm² (1.31mm² each) of combined area.

Most manufacturers use metric cabling fyi rather than AWG, since most manufacturing is done outside of the US in this industry.

 

Also what you've calculated doesn't actually make the statement untrue, it depends on what cable choices you have and are actually comparing to. If the areas were exactly the same between single and multiple the voltage drop is actually less on multiple.

 

Anyway if you go back to what I said you should include voltage stability in your thinking around this, since multiple conductors means multiple connections points meaning larger connecting surface area and more ability for the current flow to smooth out over those conductors and connections as the heat changes due to current flow and internal case temperature. 

 

Also a single 8 AWG may not be possible or practical in this application as well.

[Uttamattamakin]

On 11/9/2022 at 11:14 PM, HumdrumPenguin said:

I think every time a card is sent for RMA because of burnt connectors, the AIB or NVIDIA will replace it for the updated version that allows for an easier connection with the cable. 

Hopefully this is enough.  Making the connector easier to plug in will have to suffice unless they are going to scrap so many parts made with this new standard.  Woulnd't be the first time though many new future standards never come to be that. 

[leadeater]

18 minutes ago, Uttamattamakin said:

Hopefully this is enough.  Making the connector easier to plug in will have to suffice unless they are going to scrap so many parts made with this new standard.  Woulnd't be the first time though many new future standards never come to be that. 

I have my doubts it'll be scrapped, I think too many PSU and device manufacturers are too far down the path to change most likely. Even then revisions of the standard and guidance may be all that is required to improve the connector without much effort or significant change, quite possible these changes are exactly that.

[HumdrumPenguin]

If anyone is curious, there are many 4090 owners that can't plug in the PSU cable or adapter all the way into the graphics card's connector, myself included. It's not a matter of not enough pressure being applied, it's just a flawed design by nature. It's crazy that the most basic stuff goes by unnoticed. I don't think this is the only problem either, as there have been reports of people having melted connectors even after managing to connect the cable properly (they heard the clip click). Price aside, the card is nothing short of amazing, and this situation is frustrating, if not infuriating.

[Uttamattamakin]

Gamers Nexus has a great video on this.  They sent it to a lab and have a very thorough set of results. 

Here is a quick summary.  They sent it to a lab which sectioned a failed cable to analyze it with scanning electron microscopy.  The short findings are it is a combnination of manufacturing errors with "user error".  Error in the sense of not having inserted a badly designed connector that does not give a good click into place when it is fully seated.   

Here is my POV on this.  This is a design which runs more power, through less space.  When one does that the energy so confined looks for a path out.  Unless one engineers such a system very carefully bad things happen.  Again, my favorite analogy.  Any fool can use a well-made bottle rocket.  To launch SLS takes NASA.  It should not take NASA to plug power into a GPU. 

 

On 11/11/2022 at 6:43 PM, leadeater said:

I have my doubts it'll be scrapped, I think too many PSU and device manufacturers are too far down the path to change most likely. Even then revisions of the standard and guidance may be all that is required to improve the connector without much effort or significant change, quite possible these changes are exactly that.

This.  I can't agree more with this.  Perhaps a new generation of this connector will have a strong retention clip that will give a good CLICK when it is in place.  So that things like this ....

On 11/11/2022 at 10:15 PM, HumdrumPenguin said:

If anyone is curious, there are many 4090 owners that can't plug in the PSU cable or adapter all the way into the graphics card's connector, myself included. It's not a matter of not enough pressure being applied, it's just a flawed design by nature. It's crazy that the most basic stuff goes by unnoticed. I don't think this is the only problem either, as there have been reports of people having melted connectors even after managing to connect the cable properly (they heard the clip click). Price aside, the card is nothing short of amazing, and this situation is frustrating, if not infuriating.

... this won't be a concern.  Look at the video above.  Lots of people sort of ... jiggle the cables into place.  It should not take electron microscopy to see if you have plugged it in correctly and to be sure there are no defects in your cable.  A little scratch on the contacts, plus pulling a cable at an angle can do this. 

In short the only case you can build in with a 4090 and be sure that cable is going straight in with no kinks or pulling or anything is like a Thermaltake core W200. 

 

This is the effing case you need for that card now to be sure that cable is not bent. 

[dogwitch]

its looking more like we need to do server approce on gpu power. \

design been rock solid and can handle multi 800 watt gpus

[Albal_156]

https://www.overclock3d.net/news/gpu_displays/nvidia_hit_with_class_action_lawsuit_over_12vhpwr_cable_melting_woes/1 Someone has now launched a class action against Nvidia on the 11th of November by filing a claim in a California federal court. Idk if they will definitely win this one if Nvidia can use GNs video and say its user error though maybe NV will reduce the damages they must pay out. Maybe the plaintiffs attorneys can then say the cable was badly designed from the start (also found in GNs video) and and then nvidia will still be on the hook for this. They already said they are going to RMA/replace the cards of everyone affected but RMA is an annoying process. Wonder how big the settlement will be. Wasn't even Nvidia that designed these abominations of connectors so Jensen must be fuming.

[Kisai]

1 hour ago, Albal_156 said:

https://www.overclock3d.net/news/gpu_displays/nvidia_hit_with_class_action_lawsuit_over_12vhpwr_cable_melting_woes/1 Someone has now launched a class action against Nvidia on the 11th of November by filing a claim in a California federal court. Idk if they will definitely win this one if Nvidia can use GNs video and say its user error though maybe NV will reduce the damages they must pay out. Maybe the plaintiffs attorneys can then say the cable was badly designed from the start (also found in GNs video) and and then nvidia will still be on the hook for this. They already said they are going to RMA/replace the cards of everyone affected but RMA is an annoying process. Wonder how big the settlement will be. Wasn't even Nvidia that designed these abominations of connectors so Jensen must be fuming.

It likely would not be more than the cost of the adapter and "repair" of the GPU/PSU if they were damaged.

[leadeater]

1 hour ago, Albal_156 said:

https://www.overclock3d.net/news/gpu_displays/nvidia_hit_with_class_action_lawsuit_over_12vhpwr_cable_melting_woes/1 Someone has now launched a class action against Nvidia on the 11th of November by filing a claim in a California federal court. Idk if they will definitely win this one if Nvidia can use GNs video and say its user error though maybe NV will reduce the damages they must pay out. Maybe the plaintiffs attorneys can then say the cable was badly designed from the start (also found in GNs video) and and then nvidia will still be on the hook for this. They already said they are going to RMA/replace the cards of everyone affected but RMA is an annoying process. Wonder how big the settlement will be. Wasn't even Nvidia that designed these abominations of connectors so Jensen must be fuming.

I suspect if they focus on the cables and sockets it'll get kicked to the curb since it's an industry standard design under the purview of PCI-SIG not Nvidia. Only possible way I can see it getting past first round is evidence and argument of Nvidia controlled or contracted manufacturing defects.

 

I'd rate the success of this as extremely low.

[Deadpool2onBlu-Ray]

18 hours ago, Uttamattamakin said:

Gamers Nexus has a great video on this.  They sent it to a lab and have a very thorough set of results. 

Here is a quick summary.  They sent it to a lab which sectioned a failed cable to analyze it with scanning electron microscopy.  The short findings are it is a combnination of manufacturing errors with "user error".  Error in the sense of not having inserted a badly designed connector that does not give a good click into place when it is fully seated.   

Here is my POV on this.  This is a design which runs more power, through less space.  When one does that the energy so confined looks for a path out.  Unless one engineers such a system very carefully bad things happen.  Again, my favorite analogy.  Any fool can use a well-made bottle rocket.  To launch SLS takes NASA.  It should not take NASA to plug power into a GPU. 

 

This.  I can't agree more with this.  Perhaps a new generation of this connector will have a strong retention clip that will give a good CLICK when it is in place.  So that things like this ....

... this won't be a concern.  Look at the video above.  Lots of people sort of ... jiggle the cables into place.  It should not take electron microscopy to see if you have plugged it in correctly and to be sure there are no defects in your cable.  A little scratch on the contacts, plus pulling a cable at an angle can do this. 

In short the only case you can build in with a 4090 and be sure that cable is going straight in with no kinks or pulling or anything is like a Thermaltake core W200. 

 

This is the effing case you need for that card now to be sure that cable is not bent. 

Are you actually a  "Theoretical Astrophysicist" ?

[ewitte]

On 11/10/2022 at 12:15 AM, Kisai said:

So they changed the connector so the 4-pin is shroud is closer to the edge of the 12pin? What does this solve?

 

People not plugging it in all the way (watch the GN video).  With this change its less likely to power the card if it isn't inserted all the way.

[ewitte]

1 hour ago, leadeater said:

I suspect if they focus on the cables and sockets it'll get kicked to the curb since it's an industry standard design under the purview of PCI-SIG not Nvidia. Only possible way I can see it getting past first round is evidence and argument of Nvidia controlled or contracted manufacturing defects.

 

I'd rate the success of this as extremely low.

Plus, now there is evidence that most of this was user error...

[Uttamattamakin]

On 11/17/2022 at 9:00 AM, ShawtyT30beTHICCC said:

Are you actually a  "Theoretical Astrophysicist" ?

Yes.