Arthur T Knackerbracket has processed the following story:
Melting occurs despite Corsair's first-party 600W 12VHPWR cable being used.
Another Blackwell GPU bites the dust, as the meltdown reaper has reportedly struck a Redditor's MSI GeForce RTX 5090 Gaming Trio OC, with the impact tragically extending to the power supply as well. Ironically, the user avoided third-party cables and specifically used the original power connector, the one that was supplied with the PSU, yet both sides of the connector melted anyway.
Nvidia's GeForce RTX 50 series GPUs face an inherent design flaw where all six 12V pins are internally tied together. The GPU has no way of knowing if all cables are seated properly, preventing it from balancing the power load. In the worst-case scenario, five of the six pins may lose contact, resulting in almost 500W (41A) being drawn from a single pin. Given that PCI-SIG originally rated these pins for a maximum of 9.5A, this is a textbook fire/meltdown risk.
The GPU we're looking at today is the MSI RTX 5090 Gaming Trio OC, which, on purchase, set the Redditor back a hefty $2,900. That's still a lot better than the average price of an RTX 5090 from sites like eBay, currently sitting around $4,000. Despite using Corsair's first-party 600W 12VHPWR cable, the user was left with a melted GPU-side connector, a fate which extended to the PSU.
The damage, in the form of a charred contact point, is quite visible and clearly looks as if excess current was drawn from one specific pin, corresponding to the same design flaw mentioned above. The user is weighing an RMA for their GPU and PSU, but a GPU replacement is quite unpredictable due to persistent RTX 50 series shortages. Sadly, these incidents are still rampant despite Nvidia's assurances before launch.
With the onset of enablement drivers (R570) for Blackwell, both RTX 50 and RTX 40 series GPUs began suffering from instability and crashes. Despite multiple patches from Nvidia, RTX 40 series owners haven't seen a resolution and are still reliant on reverting to older 560-series drivers. Moreover, Nvidia's decision to discontinue 32-bit OpenCL and PhysX support with RTX 50 series GPUs has left the fate of many legacy applications and games in limbo.
As of now, the only foolproof method to secure your RTX 50 series GPU is to ensure optimal current draw through each pin. You might want to consider Asus' ROG Astral GPUs as they can provide per-pin current readings, a feature that's absent in reference RTX 5090 models. Alternatively, if feeling adventurous, maybe develop your own power connector with built-in safety measures and per-pin sensing capabilities?
(Score: 5, Interesting) by pTamok on Thursday April 10, @08:24AM (2 children)
Is there space for an adapter to fit in-line to the power connector that has six 10A fuses?
After the first pops, the rest will immediately follow, but it eliminates the fire risk. Replacing fuses is simpler than replacing connector cables and/or GPU boards and PSUs.
(Score: 4, Touché) by Mojibake Tengu on Thursday April 10, @09:02AM (1 child)
There are even electronic fuses, easily controllable. Common in industry, though obviously beyond mental horizon of NVIDIA engineers.
OEMs blindly follow NVIDIA's design, just for the purpose to stay competitive with each other.
Rust programming language offends both my Intelligence and my Spirit.
(Score: 3, Insightful) by pTamok on Thursday April 10, @09:31AM
Resettable circuit breakers and/or current limiting devices are certainly a possibility, but add complexity. An adapter with room for one traditional fuse per wire is very simple. You could probably use car 12V fuses if building one yourself from 'junk'.
(Score: 4, Informative) by anubi on Thursday April 10, @10:44AM (2 children)
We have eight glow plugs that draw 25 amperes each.
The original design has two pins in the wiring harness wired in parallel. In about ten years, enough thermal cycling / work hardening / corrosion accumulates to imbalance the current flow, and the connectors melt.
Some of us bypassed the factory connector with a heavier connector ( off-the-shelf, standard Anderson connectors ) originally designed for electric forklift chargers. I ended up buying an aftermarket 8-pin automotive connector and put each glow-plug on its own pin. On the other side of the connector, I joined all eight wires together to a single AWG#4 feed from the 200 AMP glow plug control relay. All connections are both crimped and soldered to displace water or other corrosives from wicking into the connector.
I got a bonus as being I now have eight individual glow plug circuits, it's now quite easy to isolate which glow plug circuit having issues with a standard DC clamp-on ammeter.
I guess sometimes you have to consider that sometimes a design engineer overlooks something, and to make the product usable, you have to redesign it in the field.
I used to work in Aerospace, and I know Microsoft isn't the only company who releases stuff with bugs in it. When we had those big purges of older engineers several years ago, Corporate Leadership had to experience for themselves what all those old engineers, techs, assemblers, and QA people, were so endlessly whining about. The ensuing chaos of product failures did. significant damage to the company reputation that took over half a century to build. I don't think they will ever regain the reputation they once had.
Sometimes, you just gotta make it work and fly right with what you can get.
"Prove all things; hold fast that which is good." [KJV: I Thessalonians 5:21]
(Score: 2, Interesting) by anubi on Thursday April 10, @12:57PM
Forgot to mention...an XT-60 connectors is a good place to start. They are used as battery connectors in kiddie cars, they are commonly available and come in pigtail versions of both sexes, and also PCB mount.
"Prove all things; hold fast that which is good." [KJV: I Thessalonians 5:21]
(Score: 2) by driverless on Friday April 11, @07:45AM
Also in this case it's working as intended:
It's not a meltdown risk, it's a carefully-engineered analogue-technology current-to-fire converter.
(Score: 4, Insightful) by RedGreen on Thursday April 10, @12:05PM (4 children)
So sad I can see how hard done by a moron that spent $3000 on a video card is. What a bunch of clowns paying that kind of money for a single item in a machine with get this, a known history of melting, but like so many of these types of idiots they have the but it will never happen to me syndrome, I am so special. Well cry me a river while I play the worlds tinniest violin, I have zero sympathy for fools like this and their supposed "problems" all self inflicted at this point in time as they know the garbage design they are about to buy. That is unless they are living under a rock with absolutely no contact with the world to see the literally thousands of reports of this happening.
"I modded down, down, down, and the flames went higher." -- Sven Olsen
(Score: 2, Interesting) by anubi on Thursday April 10, @01:01PM (2 children)
A lot of this kind of thing happens with cutting-edge stuff. This is a learning experience for a lot of new designers. God knows how many screwups we had trying to get the first rockets off the launch pad!
"Prove all things; hold fast that which is good." [KJV: I Thessalonians 5:21]
(Score: 5, Interesting) by stormreaver on Thursday April 10, @01:28PM
There are no mitigating factors for this level of dangerous, malicious incompetence. If NVidia wants to teach their designers how to design, it needs to be done on NVidia's time and dime. People paying $3000 for a video card are not NVidia's beta testers (the 5090 series is at more of an Alpha stage than anything that should have ever been released to the public). If NVidia wants outside testers, those testers need to be paid, and NVidia needs to supply the computer systems for testing. Charging $3000 for a video card is criminal all by itself, but charging $3000 for a shit video card that can burn down houses is a crime against humanity. NVidia should be paying people to use the 5090 series, as well as paying for health insurance and homeowner's (or renter's, as appropriate) insurance to those testers.
(Score: 3, Insightful) by VLM on Thursday April 10, @02:48PM
40 amps of current is not cutting edge for anything except for graphics cards.
(Score: 2) by bzipitidoo on Thursday April 10, @02:34PM
Coming from my habit of frugality, I was thinking something similar. Not just the initial price. There's also pouring hundreds of watts into the video subsystem. How much power does the whole machine draw, a kilowatt? And must have one heck of a beefy PSU, and lots of cooling fans. No doubt considerable additional cost to get rid of all that heat as quietly as possible.
My current desktop draws 85W max when I load it down with lots of CPU and video work. Uses just 20W most of the time, like when I'm writing or reading something.
(Score: 2) by Username on Thursday April 10, @02:49PM
Those six +12v pins all connected together as a common conductor, like the male part that's plastic, will be cast or milled aluminum instead. Drilled and filed to accept the square copper terminator. This connected by 10ga stranded wire to the psu. The same with the ground side. The two are isolated by a plastic divider between them and inserted into the plastic block with the clip on the side like it was before mod.
(Score: 2) by bart9h on Thursday April 10, @03:13PM
who cares? only suckers buy nvidia anyways..
(Score: 2) by ChrisMaple on Friday April 11, @04:20AM
A drop of mercury on each pin should ensure conductivity, at least until it eats away the pin or socket. Or until it leaks out and makes a short elsewhere.