Little PSA, derBauer is reporting that he has measured worrying temps in the 12VHPWR connector when using the 5090 Founders Edition. It seems like the card isn’t drawing evenly from all the available conductors, so one in particular is getting hot.
If you’re lucky enough to have found a 5090 FE to buy, please check the video to determine if you might be affected.
Jay from Jayz2cents has pointed out that the 12V rails in the 5090 FE go to a single busplate. He theorises that the issue is either a manufacturing problem (cables with low amps on them are not properly contacting the bus inside the card) or the PSU is not able to provide full amperage to all cables in the plug.
Both Jay and derBauer are using Corsair PSUs, but the person who experienced the burning was using an Asus Loki PSU.
I am less inclined to think PSU myself based on the initial reports… but at this stage it’s hard to tell. Allegedly there are people with third-party 5090s that do individual pin sensing that are complaining about power delivery - this would have happen before the 12V cables are bussed together - but if that part is made by nvidia then maybe it’s exposed to the same manufacturing problem? Or maybe it is the PSU somehow?
All I know is that after the last time, we should never ever have had to hear about NVidia and melting anything, especially power cables, again.
Yeah, why did they go with that crappy power connector once again?
This is extremely damning. I have always been suspicious of the connectors design as sending this much power through such a small connector can’t leave a big safety margin and it seems I was right. Someone on Reddit (yeah I know…) did a very comprehensive post about the design and power limits of the 12VHPWR cable and it’s not good. It appears that it should at best have a rating of a maximum 375W, not 600W and even that leaves less safety margin than the regular old PCIe cable standard…
Edit: I will go into more detail about the nonexistant safety margins once I am off work but basically cards like the 4090 and especially the 5090 run the cable close to the absolute maximum rating which incredibly is just a smidge over the design rating. Insanity, especially without monitoring each phase.
First the TLDR: The 12VHPWR cable should have never been allowed to hit the market as a 600W or even 450W cable. If it was designed as a 375W cable there wouldn’t be a problem as that leaves a sufficient safety margin when things go wrong but 600W is so close to the absolute power limit of the pins that things have to go only slightly out of whack in order to cause potential problems.
Here’s a very comprehensive post about the cable’s design and the underlying issues.
https://www.reddit.com/r/pcmasterrace/comments/1io4a67/an_electrical_engineers_take_on_12vhpwr_and/
The most important points:
Monitoring individual phases can’t be a fix if the phases are still bussed together on the board. It’s just a warning, if you even care to look, that there’s a manufacturing problem with your board or maybe a PSU problem… but my PSU has all of those 12V wires going to the same busbar as well (yes, through two of the older 8 pin ports)
50A 12V cable does exist of course:
Yes, however even if you have split phases like on the 3090TI, with nonexistent safety margins like running a 600W card over a single cable like the 12VHPWR the best you can hope for is that the system doesn’t turn on or limits the card to something like 150W when there isn’t enough power coming from one of the phases.
The 5090 needs a cable with proper safety margins or simply more cables with a power delivery system that isn’t run over a single shunt resistor so the card knows when there’s a bad connection somewhere. The failure points here are not the cables themselves, they do have fairly decent safety margins (still a little dicey though), it’s the connectors where things go horribly wrong. For a case like shown in der8auer’s video the card should be able to tell there’s something wrong and simply refuse to boot but you are not going to balance the load over a connector that has a problem that big (two pins running ~20A, one ~12A, others only 2A)
I only know enough about electrical circuits to minimize my chances of getting zapped.
It seems to me, that from a marketing point of view, it is better that 3% of cards ‘melt’ the cable due a dodgy connector and running a 600W card over a single cable, than it is to build in some type of protection on the card and have 90% of cards failing to boot because they detect a voltage or current issue?
NB. Percentages quoted are not based on actual data and a WAG in an attempt to make a point.
That holds until the first actual fire and everybody with a heat-damaged connector joins the class action against you 
What they could have done, of course, is design a card that actually works - the more I read the more I think this is a design fault in the card itself…
Meanwhile you have made billions and the millions in punitive damages is little more than a slap on the wrist.
Welcome to the corporate world of the 21st Century.
NB. Once again the $$$ figures are totally made up.
It’s possible, for a near term focused company.
There’s still enough places in the corporate world where reputation damage is given a lot of weight.
But yeah, given the LLM money and Nvidia’s monopoly on high-end cards, this is indeed a possibility
like 10 years later, PCI-SIG still hasn’t taken any of the initial testing feedback seriously.
we ALL told them the design is flawed, the gauge of the lines and the size of each connecting lead are too small for the current they want to be sustained through it.
I’m sure for the 126 people with 5090s this is important.
But I’m not really worried about paper catching fire.
I suspect given the demands of our hobby we’re more exposed to problems with the top-end cards than the general population
Considering an RX-9070-XT Performs on Par w/ the 4090 w/o the Power Cable Melting Drama or Fake frames…
That’s just early rumours. Let’s wait for the reviews to see how well they really perform. There could be many reasons why AMD postponed their launch.
they already said why, the drivers were horrible…
So get the hardware into vendors hands in ample supplies, optimize the drivers, see what nVidiAI is doing, release in a much better state, because drivers / software would be significantly better, and vendors would have full stock.
If AMD actually has cards people can buy, automatic win.
you know irs bad when 3rd party companies are building and selling out of 12VHPWR ans 12VPWRV2 Heatsinks, lmao.
