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

[Uttamattamakin]

Summary

The ATX 3.0 power connector aka 40 series power connectors rated up to 600 watts is in at least some cases burning and melting.  

 

Quotes

Quote

Any ATX 3.0 connector connected to a ATX 2.0 card is only half smart. 

My thoughts

Basic physics dictates that when more energy is packed into a smaller container, when the energy density is high, there will be more thermal energy released.  It does not matter if that is an ideal gas being compressed, fuel being burned, oil being pumped through a pipeline, or electricity in a wire (or a battery).  

 

Looking at this.  The bad contact Jay talks about would result in a lot of heat as all the electricity tries to push through a smaller contact. 

 

Not unlike how water speeds up when going through a narrower pipe, or shoots out of a tiny hole in a bigger pipe. 

In my educated opinion, as a physicist, there is a fundamental flaw inherent in connecting ATX 3.0 to ATX2.0.  ATX 3.0 needs that bi directional communication to be safe. That communication is what mitigates the physical issue that come with pushing more energy into a smaller, more delicate connector.   

 

Though I gladly accept being corrected by an electrical engineer.  All I know about currents is Noether's theorem.  

 

Sources

Jays 2 Cents, Buildzoid, the laws of physics.

[GuiltySpark_]

Does this video cover more than one instance of this occurring on retail 4090's? I'm seeing the same single picture all over the internet today.

[Quackers101]

10 minutes ago, GuiltySpark_ said:

Does this video cover more than one instance of this occurring on retail 4090's? I'm seeing the same single picture all over the internet today.

just an added rumor/suspicion of this and it seems like mostly the same build, could be more, which more or less an addition to the previous thread about the starting rumors of concern around this standard.

Edited October 24, 2022 by Quackers101

[Uttamattamakin]

7 minutes ago, GuiltySpark_ said:

Does this video cover more than one instance of this occurring on retail 4090's? I'm seeing the same single picture all over the internet today.

Yes there are multiple instances reported.  This one just went viral.   These are all with retail 4090's.  

The fundamental physics of the situation means there is an inherent hazard that has to be managed.  i.e. it may be the case that next generation graphics cards, at least at the high end, should only be used with ATX 3.0 power supplies.   

 

[CarlBar]

8 minutes ago, GuiltySpark_ said:

Does this video cover more than one instance of this occurring on retail 4090's? I'm seeing the same single picture all over the internet today.

 

JC in the video claims to have seen others, they just didn't blow up as big news.

 

Buildzoid also makes the point that people do all sorts of abusive thing to the old 8 pins and you barely ever hear of anything going wrong, 4090 comes out and within a very short timespan we've got a failure. Thats a worrying a trend.

[Uttamattamakin]

Just now, Quackers101 said:

just an added rumor/suspicion of this, which more or less an addition to the previous thread about the starting rumors of concern around this standard.

No this is not a rumor or suspicion.  These are pictures of one instance where what was feared would happen has indeed happened.  That is the viral one. 

This conversation reminds me of what it was like to be the first to take a SP3 into the Microsoft store due to a battery issue ... and being told that I must've done something wrong.   That was around the same year that the Samsung note 7 had it's issue.  No one wanted to think that was a design flaw until it was clear it really was. 

[Uttamattamakin]

1 minute ago, CarlBar said:

 

JC in the video claims to have seen others, they just didn't blow up as big news.

 

Buildzoid also makes the point that people do all sorts of abusive thing to the old 8 pins and you barely ever hear of anything going wrong, 4090 comes out and within a very short timespan we've got a failure. Thats a worrying a trend.

Exactly. Plus this is as I said in the OP very VERY predictable based on ... basic physics.   More energy stuffed into a smaller space, more power in a smaller pipe, means potential danger if anything goes wrong.  It is the fundamental reason that fuel requires special handling and care. 

Jay makes a great point at the end of his video.  Intel who designed the new standard (along with others) does not use it on their own cards.  NVIDIA and AMD will use the new standard.  He kinda jokes that this was Intel's master plan.   He JOKES that.  It is probably just a coincidence.  You know want a newer GPU but don't want to shell out for a PSU go for Intel.   "Intel: We won't burn up your FPS or your computer." 

[Quackers101]

3 minutes ago, Uttamattamakin said:

No this is not a rumor or suspicion.  These are pictures of one instance where what was feared would happen has indeed happened.  That is the viral one.

yeah, but we still don't know the full story. Sure its likely something about all this new stuff, but to know what actually fails in this situation can be complicated? of course the standard could make it so... it was in less of a complicated situation, where quite a few parts could be a part of the issue.

[AndreiArgeanu]

Well I guess NVIDIA's 12 or 16 pin connectors weren't that good after all and sticking to the ol' reliable 8pin might be a better idea. +1 reasons to wait to see what AMD has to provide. It's not like that many can afford 4090's in the first place or even find them. If this is/becomes a widespread issue NVIDIA may have to fork over millions at least to take every GPU in, maybe even the board partner cards but I doubt it. I can hear in the back of my head someone at EVGA going "thank god we didn't have to deal with that".

[CarlBar]

3 minutes ago, Quackers101 said:

yeah, but we still don't know the full story. Sure its likely something about all this new stuff, but to know what actually fails in this situation can be complicated? of course the standard could make it so... it was in less of a complicated situation, where quite a few parts could be a part of the issue.

 

JC and Buildzoid covered that too. The plastic at the connector is melted. The only way that happens is heat and the 4090 cannot pull more power than the connector is rated for without a power delivery failure, (which would brick the card, it's still functional despite the connector damage). There simply isn't a situation that should occur with the connector that should be able to cause this.

 

JC and Buildzoid both think that the connector wiring at some point was probably bent too close to the connector. The problem is as again both of them point out, thats normal cable management, and in a horizontal mount it's functionally impossible to fit the card and give the required unbent cable distance. Cases aren't deep enough for that, they're designed with the bendyness of the regular 8 pin in mind.And in a vertical configuration as shown in the example we now have photo's of there's often no cable management features to prevent the weight of the attached power cables themselves bending the wiring of the adapter.

[Senzelian]

First the connectors are only rated for a like 30 insertions or whatever it was.

Then the cables start melting.

Now the connectors start melting.

 

Seems to me like a garbage product. Should be replaced asap with 8-pins.

[CarlBar]

2 minutes ago, AndreiArgeanu said:

Well I guess NVIDIA's 12 or 16 pin connectors weren't that good after all and sticking to the ol' reliable 8pin might be a better idea. +1 reasons to wait to see what AMD has to provide. It's not like that many can afford 4090's in the first place or even find them. If this is/becomes a widespread issue NVIDIA may have to fork over millions at least to take every GPU in, maybe even the board partner cards but I doubt it. I can hear in the back of my head someone at EVGA going "thank god we didn't have to deal with that".

 

According to JC they'll be using the 12 pin as well. Though depending on lead time they may change that.

[IkeaGnome]

8 minutes ago, Senzelian said:

Seems to me like a garbage product. Should be replaced asap with 8-pins.

Let's take a water supply with large pipes (8 pin terminal connectors) and feed it through small pipes(12 pin has tiny terminals in it even for the main power wires) while leaving the flow the same. 

What's the worst that can happen?

[Senzelian]

2 minutes ago, IkeaGnome said:

What's the worst that can happen?

I'm sure, according to some 26yo engineer that barely knows how to design a pencil, it worked perfectly fine the 2 times he tested it in his 3million dollar lab.

 

I swear, enginners have gotten worse in the last 20 years.

[IkeaGnome]

1 minute ago, Senzelian said:

I'm sure, according to some 26yo engineer that barely knows how to design a pencil, it worked perfectly fine the 2 times he tested it in his 3million dollar lab.

Now, I'm no electrical engineer, but it seems like if someone made connectors that were rated for 150W, a person could engineer a way to plug 32 of them into the same board to give 600W. Might be able to use bigger pins. You could even split them up into convenient groups of 8.

[dizmo]

While I agree it's poor design and requires reinforcement, there's a lot of factors that are going into play that make the connector fail.

• Cards that are too large, and too close to the sides of cases, resulting in:
• Users bending the cables in ways that they aren't meant to be bent, resulting in:
• Improper contact within the connector.

They need to go back and redesign the connector. I could see this resulting in a forced recall.

The connector is technically perfectly fine for what does. In the perfectly ideal situation. That's unrealistic.

I do wonder if perhaps having 90 degree connectors would help? Then you'd have far less strain on the socket itself.

 

Prepare for more GPU shortages lads

 

51 minutes ago, GuiltySpark_ said:

Does this video cover more than one instance of this occurring on retail 4090's? I'm seeing the same single picture all over the internet today.

Loads. JohnnyGuru also commented on melted connectors, and letters were sent to the body that designed it showing issues with melting.

33 minutes ago, Uttamattamakin said:

Exactly. Plus this is as I said in the OP very VERY predictable based on ... basic physics.   More energy stuffed into a smaller space, more power in a smaller pipe, means potential danger if anything goes wrong.  It is the fundamental reason that fuel requires special handling and care. 

Jay makes a great point at the end of his video.  Intel who designed the new standard (along with others) does not use it on their own cards.  NVIDIA and AMD will use the new standard.  He kinda jokes that this was Intel's master plan.   He JOKES that.  It is probably just a coincidence.  You know want a newer GPU but don't want to shell out for a PSU go for Intel.   "Intel: We won't burn up your FPS or your computer." 

Intel's cards were supposed to be out far sooner than they were actually released, and IIRC sat in warehouses for quite a while. That's why they hadn't adopted any new connector. I imagine we will (or would have, depending on the outcome) see them on their next series of cards.

23 minutes ago, AndreiArgeanu said:

Well I guess NVIDIA's 12 or 16 pin connectors weren't that good after all and sticking to the ol' reliable 8pin might be a better idea. +1 reasons to wait to see what AMD has to provide. It's not like that many can afford 4090's in the first place or even find them. If this is/becomes a widespread issue NVIDIA may have to fork over millions at least to take every GPU in, maybe even the board partner cards but I doubt it. I can hear in the back of my head someone at EVGA going "thank god we didn't have to deal with that".

It's not an Nvidia designed connector. It's also going to be on all AMD cards.

21 minutes ago, Senzelian said:

First the connectors are only rated for a like 30 insertions or whatever it was.

Then the cables start melting.

Now the connectors start melting.

 

Seems to me like a garbage product. Should be replaced asap with 8-pins.

All connectors are only rated to 30 insertions. That's the standard.

[CommanderAlex]

I watched the Buildzoid video this morning on this and it seems the issue lies in how the cables are crimped coming from the connector. That's why cablemod says to not bend the cable horizontally in line with the connector. 

 

This crimped connection appears to maybe(? I don't know), spread apart, forming a loose contact between the cable and crimp. As a result, this makes the connection points heat up due to current flow.

[Quackers101]

3 minutes ago, dizmo said:

They need to go back and redesign the connector. I could see this resulting in a forced recall.

The connector is technically perfectly fine for what does. In the perfectly ideal situation. That's unrealistic.

I do wonder if perhaps having 90 degree connectors would help? Then you'd have far less strain on the socket itself.

agree, would wish to see something that actually focused on being usable, not "it's perfect for us".

Also I think cablemods did push some of that, with an 90 degree adapter.

3 minutes ago, dizmo said:

It's not an Nvidia designed connector. It's also going to be on all AMD cards.

as they said, nvidia wasn't the creator/designer, although did push it and earlier? Although I guess no reports for the 3090 series?

[GuiltySpark_]

52 minutes ago, Uttamattamakin said:

Yes there are multiple instances reported.  This one just went viral.   These are all with retail 4090's.  

Links? 

[Blademaster91]

This is worrying to see considering the 40 series cards and connector was released recently, it would be nice to see more reports on this, but so far it looks like GN and Jaystwocents were right on the 12 pin connector being a problem.

And this connector seems like it was either rushed too quickly, or Nvidia and the PSU companies wanted to sell more adapters and power supplies.

[CarlBar]

6 minutes ago, Quackers101 said:

Although I guess no reports for the 3090 series?

 

Power draw. Upto about 300W your basically in line with the performance of the old 8 pins on a current per pin basis. it's above that where the problem likely start to occur

 

15 minutes ago, dizmo said:

Cards that are too large, and too close to the sides of cases, resulting in:

 

This would be an issue even with older smaller cards. 35mm of unbent cable is a lot, (about 1.4 inches for you americans), you need a really deep case for that.

[xg32]

When i brought this up as the main reason i wont get a 4090 if AMD can get close, people were mostly saying the connector is fine and i was paranoid.

 

When it:

 

Can't be bent in certain ways

has small pins

uses an adapter (when there are so many of them it's bound to be an additional point of failure inside millions of computers in the future)

is designed for about ~700w max (so if 1 pin has bad contact...) this is bad engineering

even when theres no smoke or visible failures, chances are it will run hotter, and we'll see if more problems show up later down the line

 

Yes it is 1 case with pictures of exactly what was going on in the labs, but it is clearly worse as a connector than the 8pin, it's unlikely they'll recall the existing 4090s and 3090ti, but offering cards with the old 8 pin connectors max 500w shouldnt be too hard. I'll be buying an amd card and using an older psu while this plays out.

[sounds]

Here's the basic physics for what OP referenced -

Quote

Basic physics dictates that when more energy is packed into a smaller container,

Between the power supply and the 4090, adding more wires or thicker wires reduces resistance in Ohms.

 

Any resistance in the wire will reduce the voltage coming out the other end. The amount of voltage "eaten" by the resistance is Voltage Eaten = Current (Amps) * Resistance (Ohms). Then Volts * Amps = Watts of power that are turned into heat.

 

And yup, it's basic physics that a smaller wire made of the same metal must have more resistance. A bent wire will also have more resistance. A smaller connector must have more resistance. A bad connection or a little dirt will hurt too.

 

Like xg32 just wrote, this connector is way too small for a card that uses more than 400 Watts in normal operation. The good connectors are over built so that if something fails, it's not the connector melting or the wires catching on fire.

 

High voltage transmission lines get around this with ... higher voltages. The effect of the voltage drop is lessened by reducing the amps in the wire. Even then, there's the https://en.wikipedia.org/wiki/Skin_effect for AC voltages which increases the resistance due to eddy currents in the wire.

 

Maybe what NVidia needs is to ship a separate power brick just for the GPU, and use higher voltages.

[Senzelian]

30 minutes ago, dizmo said:

All connectors are only rated to 30 insertions. That's the standard.

Which standard? ATX? As in the 24-pin, EPS and PCI-e?

[Uttamattamakin]

1 hour ago, IkeaGnome said:

Let's take a water supply with large pipes (8 pin terminal connectors) and feed it through small pipes(12 pin has tiny terminals in it even for the main power wires) while leaving the flow the same. 

What's the worst that can happen?

Bingo.  

Doing that then being surprised when the pipes fail if everything isn't joined perfectly.   Not unlike a DIY'er building a computer in a standard ATX case.  Most cases people would even buy new have been on the shelf for years.  Maybe not the only case that can handle a 4090 but a good one to think about for being future proof would be the Thermaltake Core W200.  Which is more like a cabinet on wheels than a desktop case. 

 

33 minutes ago, sounds said:

Here's the basic physics for what OP referenced -

Between the power supply and the 4090, adding more wires or thicker wires reduces resistance in Ohms.

 

Any resistance in the wire will reduce the voltage coming out the other end. The amount of voltage "eaten" by the resistance is Voltage Eaten = Current (Amps) * Resistance (Ohms). Then Volts * Amps = Watts of power that are turned into heat.

 

And yup, it's basic physics that a smaller wire made of the same metal must have more resistance. A bent wire will also have more resistance. A smaller connector must have more resistance. A bad connection or a little dirt will hurt too.

 

V=IR  

P=I^2 R  

Smaller wires higher R which can translate into heat.  Just like the elements in a toaster.  Make bad contact and that point of contact will get very hot.    As I put it without math... more energy stuffed into a smaller container.  Does not matter if it is electrical energy, or pressurized gas, or mc^2 ... packing more energy into a smaller space will lead to trouble. 

IF anything a higher power card should mean bigger, beefier wires, with sturdier connections.