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

[JustBenching]

Many of users don't know of the problem until they start to smell burnt plastic. Previously I linked Tom's Hardware article about how they realized after long time that connector partially melted even though card was running fine. Imagine how many users are in similar situation that card still works but connector is already damaged.

Also, if you don't experience the issue, this doesn't mean the issue is unreal. It may still be like you don't have the issue, for now.

And how long till the millions of 4080s and 4090 users start smelling the plastic? Cause apparently it hasn't happened yet - it's been what, over 2 years...

 

[Dredi]

And how long till the millions of 4080s and 4090 users start smelling the plastic? Cause apparently it hasn't happened yet - it's been what, over 2 years...

The new one consumes more power and due to (I^2)r the increase to connectors/cables heating is exponential. Many 4090 cards have melted connectors, hundreds and hundreds of cards. Not maybe super significant amounts compared to total sales, but the problem is going to be a worse on these new cards.

In gaming, the 4090 consumes 411W, the 5090 587W, according to the TPU review. Amperage is then ~34 A vs. ~49 A. On a perfectly balanced connector that means ~5.8 A per lead vs. ~8.2 A. Assuming a great 1 milliohm resistance to each pin, the heat loss in the connectors is 0.4W on the 4090 and 0.8W on the 5090. Not a bad amount at all for either, but this is close to a best case scenario.

Assuming a maximum in spec resistance of 5 milliohms, again perfectly balanced across the connector the heat output would be 2W on 4090, but already pretty high 4W on the 5090. For reference my soldering iron in free air keeps at 300 °C at those watts.

Add some resistance imbalance and you are going to have problems for sure.

 

[JustBenching]

The new one consumes more power and due to (I^2)r the increase to connectors/cables heating is exponential. Many 4090 cards have melted connectors, hundreds and hundreds of cards. Not maybe super significant amounts compared to total sales, but the problem is going to be a worse on these new cards.

In gaming, the 4090 consumes 411W, the 5090 587W, according to the TPU review. Amperage is then ~34 A vs. ~49 A. On a perfectly balanced connector that means ~5.8 A per lead vs. ~8.2 A. Assuming a great 1 milliohm resistance to each pin, the heat loss in the connectors is 0.4W on the 4090 and 0.8W on the 5090. Not a bad amount at all for either, but this is close to a best case scenario.

Assuming a maximum in spec resistance of 5 milliohms, again perfectly balanced across the connector the heat output would be 2W on 4090, but already pretty high 4W on the 5090.

Add some resistance imbalance and you are going to have problems for sure.

If it happens on hundreds instead of millions, what's the differentiating factor? Cause there must be some. Are some cables worse quality or what? Derbauer showed that it takes less than 5 minutes for a faulty connector to get up to 140C! I can't fathom that millions of people that have 4090 haven't used their cards for more than 5 minutes at a time.

 

[londiste]

If it happens on hundreds instead of millions, what's the differentiating factor? Cause there must be some. Are some cables worse quality or what? Derbauer showed that it takes less than 5 minutes for a faulty connector to get up to 140C! I can't fathom that millions of people that have 4090 haven't used their cards for more than 5 minutes at a time.

Did you notice what connector got up to 150C?

 

[JustBenching]

Did you notice what connector got up to 150C?

The one he simulated as being faulty by having all the power go through 2 pins I think, no?

 

[londiste]

The one he simulated as being faulty by having all the power go through 2 pins I think, no?

Yes. But the 12V-2x6 was 87C or around there. 150C was on the PSU side what looks like an 8-pin - the more sturdy one that PSUs push 300W through.

 

[JustBenching]

Yes. But the 12V-2x6 was 87C or around there. 150C was on the PSU side what looks like an 8-pin - the more sturdy one that PSUs push 300W through.

You sure it was an 8pin that pushed to 150C? I thought it's a 12vh power cable on both sides.

 

[londiste]

You sure it was an 8pin that pushed to 150C? I thought it's a 12vh power cable on both sides.

14:39 and 14:56 in the video:

This is the PSU he uses:

AX1600i Digital ATX Power Supply — 1600 Watt Fully-Modular PSU

The CORSAIR AX1600i is the ultimate digital ATX power supply, built using only the best components and state-of-the-art gallium nitride transistors to deliver more than 94% efficiency.

www.corsair.com

I am absolutely not trying to say there is not a problem but I am getting a feeling that the problem is a bit more interesting than the 12V-2x6 connector

 

[Dredi]

Are some cables worse quality or what?

Even an in spec cable with perfect current balancing can heat up at 4W (inside the connector).
Some are going to be worse quality for sure, but the problem is with the spec, combined with the nvidia card implementation. The increased power consumption alone will cause there to be a lot more problems now.


Broken cables cause problems on any system, even those that are not using the 12 pin connector.

Derbauer showed that it takes less than 5 minutes for a faulty connector to get up to 140C! I can't fathom that millions of people that have 4090 haven't used their cards for more than 5 minutes at a time.

That cable was faulty. It’s super easy to verify that by just calculating the minimum resistance for each lead that would cause the measured current values.

A good demonstration on the fact that the psu and gpu have no idea and no control of what’s going on between them.

 

[Werwulf]

The solution is simple, just relocate to a zero-G environment and float your cables out in a straight line from the GPU. Hope you didn't want to close your side panel.

 

[Dredi]

The solution is simple, just relocate to a zero-G environment and float your cables out in a straight line from the GPU. Hope you didn't want to close your side panel.


View attachment 386065

Someone should make a new connector with 35mm extra length. Problem solved /s

 

[Bobaganoosh]

If it happens on hundreds instead of millions, what's the differentiating factor? Cause there must be some. Are some cables worse quality or what? Derbauer showed that it takes less than 5 minutes for a faulty connector to get up to 140C! I can't fathom that millions of people that have 4090 haven't used their cards for more than 5 minutes at a time.

Where are you getting that Nvidia sold "millions and millions" of 4090s? The number is probably a few hundred thousand globally. The issue with this connection is only when pushing it hard, for the most part, so it's mostly even just the people running their 4090 over 450W, which is likely not even a huge percentage of those owners. It seems as though they've only sold about 100 5090s so far, so we'll see much higher percentage of fallout there. The issue is more likely to occur than most people would find acceptable due to very low safety margins and almost no precautions taken on the design of the cables or cards, but that doesn't mean every single cable/card is a ticking time bomb. It's not always "all or nothing" lol.

 

[JustBenching]

Where are you getting that Nvidia sold "millions and millions" of 4090s? The number is probably a few hundred thousand globally. The issue with this connection is only when pushing it hard, for the most part, so it's mostly even just the people running their 4090 over 450W, which is likely not even a huge percentage of those owners. It seems as though they've only sold about 100 5090s so far, so we'll see much higher percentage of fallout there. The issue is more likely to occur than most people would find acceptable due to very low safety margins and almost no precautions taken on the design of the cables or cards, but that doesn't mean every single cable/card is a ticking time bomb. It's not always "all or nothing" lol.

Well according to some reports there were 160k 4090s sold during the first month, so I assumed over the 2 year span it would reach million (s). Dunno, maybe those numbers were off.

 

[Werwulf]

Someone should make a new connector with 35mm extra length. Problem solved /s

You'd think if you thought your cable couldn't tolerate any bending close to the connector, you'd put some kind of sheath or strain relief on it but then nobody would buy it because you wouldn't be able to close most cases 35mm plus any bend radius is what, at least 50mm? Who has that between a big AIB GPU and their case side panel??

 

[taka]

You'd think if you thought your cable couldn't tolerate any bending close to the connector, you'd put some kind of sheath or strain relief on it but then nobody would buy it because you wouldn't be able to close most cases 35mm plus any bend radius is what, at least 50mm? Who has that between a big AIB GPU and their case side panel??

In a 7+2 or 7+3 PCI brackets case there is enough space for some cards. Side panel is closed.

3090 TUF vs 4090 TUF also as comparison.

 

[Werwulf]

In a 7+2 or 7+3 PCI brackets case there is enough space for some cards. Side panel is closed.

Yeah, that's a big case. Is that a 3080 TUF? A 5080 TUF is even wider. I'd like to see GPU manufacturers standardise their width measurements to the tip of the power connector, at least that way you'd have some chance of knowing how close the cable is going to be on your new GPU *before* you buy it!

 

[Dredi]

In a 7+2 or 7+3 PCI brackets case there is enough space for some cards. Side panel is closed.

3090 vs 4090 also pictured

Those leads are not straight though. Clearly they bend down almost immeaditely from the connector.

 

[N/A]

or this.

 

[taka]

Those leads are not straight though. Clearly they bend down almost immeaditely from the connector.

I just opened the case and here is the situation. The angle of the shot is different.

 

[ShrimpBrime]

As long as the wire doesn't have a sharp link, it just doesn't matter unless the gauge is too small.

 

[Bobaganoosh]

Well according to some reports there were 160k 4090s sold during the first month, so I assumed over the 2 year span it would reach million (s). Dunno, maybe those numbers were off.

I saw that, but it was also reported that supply dried up substantially after that and prices skyrocketed after that, dropping sales. I'm not saying they sold none, but I was just wondering where the "millions" came from.

 

[taka]

As long as the wire doesn't have a sharp link, it just doesn't matter unless the gauge is too small.

6 times the diameter of the cable equals the radius of the bending, for most cables

FAQ: Determining the minimum bending radius | Eland Cables

Answering the frequently asked question: how is the minimum bend radius determined for electrical cables. The minimum bending radius is a measure of the smallest radius a cable can be bent around without damaging the cable. Discover the factors that influence it.

www.elandcables.com

 

[ShrimpBrime]

6 times the diameter of the cable equals the radius of the bending, for most cables

View attachment 386100

FAQ: Determining the minimum bending radius | Eland Cables

Answering the frequently asked question: how is the minimum bend radius determined for electrical cables. The minimum bending radius is a measure of the smallest radius a cable can be bent around without damaging the cable. Discover the factors that influence it.

www.elandcables.com

For a wire that small, the bend is sharp anyways.
Got it. Thank you!

 

[Werwulf]

Jon Gerow (i.e Corsair) thinks it's ok to bend close to the connector, just need to check you haven't popped any pins out.

 

[Dredi]

I just opened the case and here is the situation. The angle of the shot is different.

BEND!