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

[BoggledBeagle]

So do only the 3rd party and Corsair cables burn?

Any cable with damaged or poorly manufactured plugs will have problems.

It seems that there are two different designs of the internal connector tubes - the original one slot and current two slot tubes, which probably have better grip on the pins and are more resilient.

I would not trust the original plugs even with a limited number of insertions behind them.

What was your test setup? Why not test with 600W?

Follow the link to the survey thread above. I have only a card that can draw 400W.

 

[Dredi]

Follow the link to the survey thread above. I have only a card that can draw 400W.

Oh, ok. For cable tests you should follow the aphenol testing procedure:

https://cdn.amphenol-cs.com/media/wysiwyg/files/documentation/gs-12-1706.pdf

30 connection cycles etc.

 

[BoggledBeagle]

Oh, ok. For cable tests you should follow the aphenol testing procedure:

https://cdn.amphenol-cs.com/media/wysiwyg/files/documentation/gs-12-1706.pdf

30 connection cycles etc.

Why? The thread is meant to try to identify problematic products by people posting what cables exactly they used, how they used them (including how many times they plugged them in) and what were the results.

I suspect the thread will not have many inputs, because people would need to admit, how much they abused the cables before they suffered some damage.

 

[Dredi]

Why? The thread is meant to try to identify problematic products by people posting what cables exactly they used, how they used them (including how many times they plugged them in) and what were the results.

I suspect the thread will not have many inputs, because people would need to admit, how much they abused the cables before they suffered some damage.

Well, if you’d want to find out if a cable is ”problematic”, should you not test it with the correct methodology for determining that? Your tested cable can be just fine for single connection cycle, but fail at 10 cycles. Who knows. There are also environmental tests which would indicate if a cable could fail over time in moist sea air etc. Such a cable would definitely be problematic, but your simple test will not find that out.

your test is too simple to be useful.

 

[BoggledBeagle]

Well, if you’d want to find out if a cable is ”problematic”, should you not test it with the correct methodology for determining that?

I do not run a lab and have no budget.

 

[lilhasselhoffer]

The manufacturer want to sell products in the specific price/perf range, that's understandable. But if you watch Buildzoid's video you'll see he explains it very well, each new generation 16 pin gets weaker on the GPU side. If you say a multi-billion dollar company doesn't have the experts to understand this and they do it because they are fools, and if you said that the consumer must accept that like something normal, I don't think so

About binning, a lot records in 3DMark are made with 700w+ mods, so the chips/cables/connectors are not the problem here.

I know why i do not read such posts and responses.

No and No. Definitely not.

I think you misunderstood a good design principle.

You do that on the hardware level - on the component level in the firmware. Not in the userspace software.

There are some sort of microcontrollers. These are logic with functions and flash memory. That firmware and that circuitry should limit the current flow. Assuming you are not a "smart guy" who desinged that circuit. You have those parts in any power supply unit, television set, and so on.

someone was lazy designing that power circuit.

#719 - That guy is also right. Shunt mods and such. 700Watt or more over the same connector. I have a psu with the older nvidia gpu connector and i have two psu with the usual amd graphic card psu connectors. 600 Watts over such small connectors in comparision with the other connectors - definitely not to my liking.

#719 I would not give a free pass for every chips/cables/connectors

-- I have to admit something is wrong. My conclusion is a do not buy badge over all Nvidia related products.

Anyway I'm lucky that I was not affected because I did not buy Intel 13th or 14th gen processors or a nvidia graphic card.

It's sad still no solution or statement from NVIDIA. I want a real statement and no bullshit statement. I want to see actions. Real actions which solves this permanently.

NVIDIA Claims 16-Pin Power Connector Issues are Over, No More Melting

During a recent press event in South Korea, NVIDIA addressed concerns about power connector safety for their upcoming RTX 5090 graphics card. The new GPU will consume 575 watts of power, marking a massive 225-watt increase from its predecessor, the RTX 4090. The previous generation RTX 4090...

www.techpowerup.com

This was bullshit and easy to see bullshit by nvidia



Nicely Whataboutism: Intel can pull around 460 Watt permanently to the cpu socket on consumer mainboards. No issues with cables on those mainboards. A big increase. MAybe it is valid to compare a intel cpu with 460 Watts with a NVIDIA 4090.

You seem to think I'm defending Nvidia...because? I am not. What I'm saying is that they decided to put out a product based off of a cheaper connector interface. In the process they did a stupid thing.

To quantify that stupidity, they decided to introduce an electric circuit that has a specification that is not well thought out. It was then compounded by designing hardware that did not have protections. The summation of the entire situation is and was that they saved money with less interface and hardware...which is really stupid because it both removes protections and components.


A proposed solution would be to use a few cents of shunt resistors, which would allow you to have a watchdog style circuit that could manipulate the power requested based upon the measured amperage in each conductor...which I didn't ever say had to be transparent to the user. It'd be entirely reasonable in fact to make this a part of internal processes that users had no visibility to. Likewise, a shunt resistor monitored for amperage could also be blown...which is why it is less ideal than a breaker which can toggle should it pull too much power through it (thus requiring no monitoring and being an automatic response rather than a detection). I also stated that this sort of thing cannot influence the severity of a failure, but a detection that was automatic would mathematically make-up for a higher failure rate...allowing for said few cents of electronics to make all of this disappear.



I think you want to believe that I suggested usage of shunt resistors similar to overclockers...who remove all internal protections and then manually manipulate power settings. That's fine and dandy, but not a good way to sell consumer products. I'm stating, flatly, that the design of the connector and its surrounding protections was made by somebody in a lab setting, that did not verify it in real world settings. In the real world the removal of internal monitoring, along with silly high draws which are not balanced in any meaningful way, is the failure point being observed. My criticism is Nvidia could have undone this with a few cents of hardware, but because their cheap new connector spec was fine on paper they have painted themselves into a corner where they either have to take a third look at the spec, stop pulling power in insane amounts, or add back in some of the costs "they didn't need" to be safe again. The only alternative is to install multiple x8 connectors...like AMD is already doing and Nvidia used to do before they decided to save money.

We cool now? No questions? No loyalties to brands, because that's stupid. When you design a spec and don't test it in the real world conditions that will exist you generally either have to be dang sure it won't fail (like 1 in millions of test cases sure), or you have to build in protections that will detect and respond to failures without the user whenever your failure mode is capable of immediate and serious bodily harm. Nvidia did neither of these things, and their connector shows that they only met an internal goal of cost mitigation.



Side note, this is why we have things like a test where to prove that your fuel tank is good they specify how far into water they must be submerged and at what temperature the water is. They developed this requirement because one operator wanted to get around the tests so they met the requirements by having a 2" deep tank...that barely even got the tanks wet. When called on that, they decided to sample the tanks by filling them, freezing them, and then submerging them so that any imperfections were sealed with ice. It sounds really stupid to have a dunk test with three paragraphs explaining how to dunk a tank into water and watch for bubbles...but sometimes engineers cannot account for other engineers and reality. In this case, and on paper, a card that pulls 575 watts should have no problems with a 600 watt potential source...right?

 

[BoggledBeagle]

I'll never understand the pathological urge some of you have to speak up for a megacorporation that refuses to change a design that's been proven to introduce serious problems for the end user.

How?

Use with damaged cables, incorrectly installed or incorrectly manufactured cables is irrelevant.

Provide a link to a single case of a correctly installed, undamaged and well made cable that caused a problem. It should be well documented and reliable.

 

[Dredi]

Provide a link to a single case of a correctly installed, undamaged and well made cable that caused a problem. It should be well documented and reliable.

How do you determine that all your requirements are met?

 

[BoggledBeagle]

How do you determine that all your requirements are met?

It will be easy, if the case is well documented.

 

[Maxiaid]

How?

Use with damaged cables, incorrectly installed or incorrectly manufactured cables is irrelevant.

Irrelevant? This is a product for the DIY market. It should be pretty much idiot-proof and feature big safety margins. 1.1x is just ridiculous and irresponsible engineering. You have to know the use case you're marketing for.

Provide a link to a single case of a correctly installed, undamaged and well made cable that caused a problem.

There's been countless reports of people who had their cards running "just fine" for over a year before their connector melted. And your definition of "damaged cables" is microscopic wear of the coupling surfaces that may even be happening insidiously over time, not exactly a rusty cable with exposed wires. This whole conversation is just ridiculous. A USB-C cable, for example, can be mated hundreds of times, no problem. And this crappy solution for a premium $2000+ GPU can't take even 30? If you're gonna be running that much power through a connector you must make sure it will work reliably. I really can't grasp your siding here.

 

[Dredi]

It will be easy, if the case is well documented.

What specifically needs to be documented for you to approve?

 

[BoggledBeagle]

I will be happy to assess the veracity of the case, once it is available.

It really does not make sense to talk about something that does not exist.

 

[Dredi]

I will be happy to assess the veracity of the case, once it is available.

It really does not make sense to talk about something that does not exist.

Unless you give specific requirements this is pointless, as you will simply respond to any and all cases with ”they did not have documentation on XYZ, which makes this invalid”.

 

[Vayra86]

No one makes PTC thermistors for balancing 8Amp steady state loads across 6x pins. Said thermistors are closer to self resetting fuses. When those things pop their resistance climbs to the point where the circuit is effectively cut off.

Copper naturally rises in resistance the hotter it gets. Parallel copper lines, in theory, are therefore self balancing to some degree.

It makes you wonder what the problem is here exactly.

I'm not really buying the 'but load balancing fixes all things' line either - never did. It still comes down to a simple observation: flimsy POS cable, too many watts.

We never talked about load balancing on pcie either, not even on 400+W GPUs running over 3x8 pin, and the reasons are again... super obvious... more separate connections to divide the load through straight from the PSU... and far higher tolerances so even if 1x8 pin completely fails, you could still run the hardware just fine - unless it won't even run if one of the connectors isn't providing juice. These kinds of failsafes are impossible to get around, and 12vhpwr has none of them.

Page 30. Just 20 to go...

 

[BoggledBeagle]

Irrelevant? This is a product for the DIY market. It should be pretty much idiot-proof and feature big safety margins. 1.1x is just ridiculous and irresponsible engineering. You have to know the use case you're marketing for.

Building a computer is not idiot proof. You need to be handy, have specific knowledge and skill.

There's been countless reports of people who had their cards running "just fine" for over a year before their connector melted.

So what? Was their cable OK? Was it correctly installed?

And your definition of "damaged cables" is microscopic wear of the coupling surfaces that may even be happening insidiously over time, not exactly a rusty cable with exposed wires.

You can pretty easilly cause macrospopic deformation of the connector tubes by incorrectly inserting or deforming the plug while it is inserted. I wrote in the survey thread that the plastic plug body felt soft. If you were exerting a force on this plug by a cable, it could bend or twist in the socket.

A USB-C cable, for example, can be mated hundreds of times, no problem.

So what? It is a different connector made for different purpose. Mentioning it is irrelevant.

 

[Vayra86]

Which part of "it's a cable and/or connector problem" do you not understand? "Redesigning the PCB" isn't going to change that.

Buy a good quality cable and don't abuse it. You'll be fine.

You're being awfully short sighted here. How do you fix this with the same cable that is the problem?

Its definitely an Nvidia problem because they give you a product that pulls enough juice to destroy this cable. Its also a cable they chose to place on this GPU, and they even chose to place just one connector for it, so you're screwed no matter what cable you get. It can fail at any given point in time, you're just praying it won't.

But if your totally randomly concluded 'You'll be fine' gives buyers peace of heart, power to them

 

[Dredi]

I'm not really buying the 'but load balancing fixes all things' line either - never did. It still comes down to a simple observation: flimsy POS cable, too many watts.

We never talked about load balancing on pcie either, not even on 400+W GPUs running over 3x8 pin, and the reasons are again... super obvious... more separate connections to divide the load through straight from the PSU... and far higher tolerances so even if 1x8 pin completely fails, you could still run the hardware just fine - unless it won't even run if one of the connectors isn't in there. These kinds of failsafes are impossible to get around, and 12vhpwr has none of them.

It would fix all of the cases where the cable is within or close to spec.

It would not be able to safely use high resistance connectors, or badly connected cables, as they would still melt due to contact resistance heating things up. For that the only solution is to add more surface area for heat dissipation in relation to aperage of the connectors. The latter however will not be enough to solve all problems, and load balancing is still required.

adding more connectors solves nothing unless load balancing is applied.

 

[Vayra86]

It would fix all of the cases where the cable is within or close to spec.

It would not be able to safely use high resistance connectors, or badly connected cables, as they would still melt due to contact resistance heating things up. For that the only solution is to add more surface area for heat dissipation in relation to aperage of the connectors. The latter however will not be enough to solve all problems, and load balancing is still required.

Nah you just need to slice the max wattage allowed on these cables by 2 or 3. Simple fix, except it won't help the 5090. That product is beyond saving, at least the FE and all other variants with a single connector. You will heavily undervolt it, or you'll eventually simply see it fail.

 

[Dredi]

Nah you just need to slice the max wattage allowed on these cables by 2 or 3. Simple fix, except it won't help the 5090. That product is beyond saving, at least the FE and all other variants with a single connector. You will heavily undervolt it, or you'll eventually simply see it fail.

And you need to balance the load between them…

 

[Vayra86]

And you need to balance the load between them…

Why? You're already limiting each cable to a hard 200W. If the resistance in one of them is too high, there's enough hardware to move that load elsewhere.

 

[Dredi]

You're already limiting each cable to a hard 200W.

How? With load balancing you could.

 

[Vayra86]

How? With load balancing you could.

Isn't this already happening automagically because you now have multiple separate connections to the PSU? Why would the PSU funnel the full 600W through one cable and nothing through the rest?

 

[Dredi]

Isn't this already happening automagically because you now have multiple separate connections to the PSU? Why would the PSU funnel the full 600W through one cable and nothing through the rest?

Without load balancing on a single rail power supply that could still happen if the connectors on the gpu side are not load balanced.

on a typical multi connector gpu the gpu does load balancing to ensure that the 150W limit of each 8pin is not exceeded. Or 75W of a 6 pin.

the psu CANNOT soft limit the amount of power given to any pin/connector, because it has no such capabilities. It can only trigger OCP and shut down.

if no load balancing would be applied the current would skew, same as this 12pin nividia shit.

 

[Vayra86]

Without load balancing on a single rail power supply that could still happen if the connectors on the gpu side are not load balanced.

on a typical multi connector gpu the gpu does load balancing to ensure that the 150W limit of each 8pin is not exceeded. Or 75W of a 6 pin.

if no load balancing would be applied the current would skew, same as this 12pin nividia shit.

K. Didn't know that; but like you say... should come with the connectors on the gpu then.

 

[Maxiaid]

Building a computer is not idiot proof. You need to be handy, have specific knowledge and skill.

Man, people with over a decade of experience with building computers have had the problem, stop giving yourself airs. The design lacks failsafes. That is called poor engineering. End of story.

So what? It is a different connector made for different purpose. Mentioning it is irrelevant.

I can see what your favorite word is. I stand by what I said.