Another Nvidia RTX 5090 cable melts despite MSI's "foolproof" yellow-tipped GPU connector

[user556]

Given the original problem is fading away I'd say the crappy cable plug pins have been identified and removed from the supply chain. These MSI ones might be something of an exception ... that hopefully they'll get sorted too.

That just leaves the still underused cables out in the wild as a future time bomb for those that might later use a flawed cable with a high power GPU.

There really should be an industry recall removing the bad cables from the wild. If only to put an end to all the speculation.

 

[BogdanR]

I posted this already, here it is again:

The best cables I've seen are 16 AWG. A simple calculation for a solid 16 AWG wire shows that if fed 12V at 8A at the PSU end, it will deliver 11.9 V at the other end if it is 3 ft long (standard). However those wires are not one single core solid wire but multicore so skin effects will drop the voltage further. This will make the card ask for slightly more current to compensate, that's the law and that is your design flaw at least partially.

I'm currently holding one of those 12VHPWR cables proudly displaying a 600W label on it and it is made out of 16 AWG wires which are inadequate for this sort of current draw, their safety factor is simply too low. To be more precise, if they would be a few inches longer we would have more than 1% voltage drop. Sure, such a wire will take 9A draw individually but not without heating up, more so due to skin effects.

And then you look at the actual connector's sizes, the actual bit which has to transmit those 8 amps, it is simply too small, it barely has enough contact area for that current capacity if everything is ideal and the contact is almost perfect. Reality is different, contacts may not be perfectly clean or have a tiny bit of oxidation on them which is not uncommon, so they will heat up when the current draw increases, causing a bit more oxidation and so on, until the temperature caused by the increased resistivity of the contact melts the connector.

As an Automation Senior Engineer I'm quite familiar with power requirements and how power distribution works. When connectors melt could be due to short term peak current draw but most often it is because the RMS current draw is beyond what the connector's duty cycle can take. Therefore the only logical conclusion one can take is the wires/ connectors are undersized. The hard evidence points that way.

 

[user556]

... but most often it is because the RMS current draw is beyond what the connector's duty cycle can take. Therefore the only logical conclusion one can take is the wires/ connectors are undersized. The hard evidence points that way.

The hard evidence is the problem has transitioned from a statistically small percentage of installs to a vanishingly small percentage. At least in terms of ones installed with a heavy load. The way I see it is we have always been looking at quality of the cable plug pins, not any flaw in the spec.

 

[toooooot]

Everybody is trying to save Nvidia's failed design.
I hope they discuss it with other companies before releasing it.
They are not the company that is best at making cables and connectors.
In fact, they might be a terrible one in this department.

 

[Squid Surprise]

Everybody is trying to save Nvidia's failed design.
I hope they discuss it with other companies before releasing it.
They are not the company that is best at making cables and connectors.
In fact, they might be a terrible one in this department.

They have nothing to “save”… the problems have mostly disappeared and there is nothing that proves it was ever Nvidia’s fault to begin with… meanwhile, they sell gazillions of them every year…

 

[BogdanR]

The hard evidence is the problem has transitioned from a statistically small percentage of installs to a vanishingly small percentage. At least in terms of ones installed with a heavy load.

When the installed base is small, the number of these fault occurrences becomes more statistically significant, not less.

The way I see it is we have always been looking at quality of the cable plug pins, not any flaw in the spec.

The connection could use a more robust design with more current carrying capacity. The same goes for the rather marginal section cable. The next size up would be most likely adequate.

 

[Axeia]

The yellow-tip idea from MSI is actually smart in theory, but it's clearly not a fix-all. Between tight cases, stiff cables, and GPUs the size of small countries, "fully seated" can mean different things depending on your setup. Maybe it's time for a complete connector rethink instead of color-coding band-aids.

That'd be in large part up to NVIDIA. As I understand it board partners aren't allowed to for example stick the good old 6+2 pin on lower power models.

AMD seems to allow a little bit more freedom in that regard and there's some AMD cards with the dreaded 12VHPWR connector.

I guess with how long the old 6+2 pin lasted this is not something we want to be changing very often and we're stuck with this imperfect successor for quite some time.