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

[Dredi]

Well one example i know of for sure is that Asus Thor PSUs have it for their 12VHPWR connector.

What does it actually do though? I wasn't able to find any review where any such feature would have been introduced, or tested.

 

[_roman_]

There are already Asus Astral owners getting annoyed because the software is essentially telling them they need to replace their cable because even a small imbalance puts a pin in the red.

And people believe these "uncalibrated" values. That is not a measurement device. check the paper manual of any decent true rms multimeter or true rms clamp multimeter.

Note: Well I aid in the goal to achieve 50 pages. 41 pages are not enough.

 

[Dredi]

And people believe these "uncalibrated" values.

Are you certain that Asus has not calibrated them as a part of the cards production? Unlikely yes, but possible.

What has been the measured delta between what the app says compared to a calibrated clamp meter? 1%?

 

[lilhasselhoffer]

Real cause is already known.

You can apply primitive NxWhy method (basic analyzing method that 3-4 years old children apply to test their parents patience).

Connector specification (design) is the main problem here IMHO. It was created with almost no safety margin in mind. It's living on the edge. Many of you here think the solution would be to monitor per pin current and adjust current to be evenly distributed among pins. Lack of this feature is not real root cause here, that is way how to workaround current problem. Of course, current distribution control increases safety a lot, but had the pin resistances not vary that much, connector would be fine even without workarounds in place.

Yeah if you added a safety mechanism to all 5090s to shutdown when a pin goes out of spec, people's cards would be shutting down all the time

There are already Asus Astral owners getting annoyed because the software is essentially telling them they need to replace their cable because even a small imbalance puts a pin in the red.

There's two ways I see of answering the failure. It's funny, because the inflection point between them is whether you look at this as a resistance issue or a current draw issue.

Define problem: 12 pin connector cards are burning up.

Why are they burning up?
The resistance in the connector, combined with the current flow, causes them to heat up.
Why is the resistance causing heating?
Connectors are receiving too much power, and their resistance converts too much energy to heat.
Why is the resistance so high?
The cross sectional area of contact is too low for the current flowing
A -> increase the cross sectional area of the connector

Why are they burning up?
The resistance in the connector, combined with the current flow, causes them to heat up.
Why is there so much current flow
The card is pulling near specification limits to power the GPU
Why is the specification limit so close to the draw?
Nvidia designed the connector assuming an even distribution of current.
Why did they design assume that the current would be evenly distributed?
They didn't understand the difference between practical application and theoretical load balancing
A -> They need to install a management system to prevent unbalanced current loading on the connector.


The two logical answers, depending upon how you write the question, is to either dramatically increase the cross sectional area (and thus decrease the resistance of the connection) or to install current limiting to the system so none of the lines exceeds the energy required to incinerate the connector. Note that the later does not specify whether you choose something like a fuse, breaker, active monitoring and management, etc...

The endgame though is the same. The connector and draw combination mean that your specification does not match your requirements...and how you define why it doesn't meet your purpose leads you to how you answer the question. I'm sure a manufacturer would want active monitoring, given that absolves them of issues with minimal cost. I'm sure that the software guys would want a beefier connector, because that metal isn't worth a fraction of what their coding is...at a one time expense. As consumers, we look at this as Ngreedia because both solutions will take less than a few dollars to implement...and they decided to cheap out while dramatically increasing the price of a card.

 

[Werwulf]

And people believe these "uncalibrated" values. That is not a measurement device. check the paper manual of any decent true rms multimeter or true rms clamp multimeter.

Note: Well I aid in the goal to achieve 50 pages. 41 pages are not enough.

If your 5090 is pulling nearly 9.0A through each cable and your dog farts in the next room, you're probably going over those limits.

 

[lexluthermiester]

I just know we're gonna hit page 50 soon. 50 pages of copium fighting dosages of reality.
I can almost smell it! Hmmm. Burnt plastic!

Let's face facts, the whole damn situation is unacceptable. We need a better solution.

 

[freeagent]

How come all the plugs that burn only have 2 wires?

That one has 4 wires.. why are people using the other one?

 

[N/A]

CARD_PWR_STABLE can provide a fault detection alert from the Add-in Cards to the PSU, which can provide the PSU an opportunity for protective measures.

CARD_CBL_PRES provides a constant DC logic signal from the Add-in Card to the power supply to indicate that the 12V-2x6 Connector is correctly attached.

What if PSU still doesn't do anything about it. That happened in the first post. all four also.

 

[x4it3n]

3080 Ti FE, nothing melted with OEM (Seasonic) 2x8pin to 12-pin cable, been going strong for 2 3 years (or is it 3 now, can't remember Jesus, it's been 3 years already, time sure flies).

3090 Ti was the one that came up with 12VHPWR but it wasn't until 4090 when the melting started. 40 series cards that use standard 8-pin connectors are lower tier with much lower power draw, plus we haven't really seen any melting on something other than a 4090 before the 50 series showed up.

Yeah the 3090 Ti didn't have that issue apparently. Also I wonder what people with burning connectors are doing...or if they are defective units/cables/connectors, because I have a 4090 SUPRIM LIQUID X and even with OC + 530W BIOS it's not melting. Not that I want it to melt but so far no issues. Also I'm using the 4x 8-pin to 16-pin adapter. But I bought a Seasonic Prime TX-1600 Noctua Edition (ATX 3.1 & PCIe 5.1) on purpose so I can use their 12V-2x6 cables on a 5090 when I get mine this month (if everything goes well).

I think that this much focus on connector is misplaced. Yes, it had teething problems and there is precious little safety margin compared to old 8-pin but is that the cause of the meltdowns we are seeing? The 12-pin on 30x0 was effectively the exact same design and these did not melt in any significant amounts. There were a few examples that largely did boil down to user error and big splash in hardware news. The real problem started with 4090 and it wasn't exactly clear, why. 3090Ti was running the same 450W through it and 3080Ti/3090 with 350W were not really that far off.

The 3090 Ti was using 3 shunt resistors to divide current evenly on all cables. The 4090/5090 just have 1 or 2 depending on models.

 

[londiste]

If your 5090 is pulling nearly 9.0A through each cable and your dog farts in the next room, you're probably going over those limits.

600W at 12V is 50A. 8.3A for every wire. Limit in the spec is connectors at 9.3A each. Yes, that only leaves a very small 12% margin.

The 3090 Ti was using 3 shunt resistors to divide current evenly on all cables. The 4090/5090 just have 1 or 2 depending on models.

Shunts themselves do not divide, they with (usually) help measuring the current. There need to be some circuit or the power delivery design itself that deals with evening out load on input.

What if PSU still doesn't do anything about it. That happened in the first post. all four also.

From what it looks like on PSU side it is also a big blob of 12V and PSU does not and by definition should not care other than providing 12V power and ideally keeping within the limits requested/allowed by what card says through sense wires. CARD_CBL_PRES should help if the connector is not properly in. If it is properly in the uneven distribution of power over wires can still happen and melt things.