Nvidia faces lawsuit for melting RTX 4090 cables as AMD has a laugh A lawsuit seeking class-action status has accused Nvidia of misleading consumers over the safety of the company's GeForce RTX 4090 graphics cards due to growing reports of melting cables. The lawsuit, filed on November 11 by Lucas Genova in the US District Court for the Northern District of California, looks to charge Nvidia …
Gamer's Nexus on YT recently put up a video where they sent a number of failed connectors to a failure analysis lab, in all cases the lab thought the fault was down to incorrect installation due to not seating the connector correctly. So the lawyers may get short shrift if these things are looked at, at microscopic scales.
Though there certainly is an argument to made that the connector lacks feedback such as a positive click or visually confirmable mechanism for the user to validate the connector is correctly inserted.
Even more argument could be made that Nvidia might have been negligent in putting the connector on the outboard side of a card that is difficult to fit in the most generous case without excessively bending the power cable and placing strain on the connector.
I just watched the video... it"s not all cases and it seems it's a lazy design. The video runs just the cable angle, there's no mention of testing if the GPU circuit or software can be faulty... nothing not a harness test :-/.
I understand for flexibility that they want to keep the lead and thread count high, but if you have to go thicker who cares as It's a 1 time install. Throw some 10awg at it... it'll be alright.
Forgive me, I may have missed the bit where Steve caveated the all cases with 'apart from these ones with FOD, and from the other connector supplier'.
I was dealing with exactly how I was going to correlate 2 customer datasets. Where in one of them, the system supplier appears to have decided that 'input validation is for the weak', and the users have discovered the forename field is a great place to put freeform notes. It has not been a good week.
I can assure you, I agree wholeheartedly with you regarding the poncy conductor thickness/cost-cutting/connector walled garden strategy, and I for one will not be purchasing a GPU where the power connector has, apparently, a 40 insertion lifespan.
So, as Apple would say, they are inserting it wrongly? I, and i'm sure many of us, have built computers over the years, our first one will have been, to put it bluntly thrown together with cables and cards plugged in without much care. A fire shouldn't occur because someone wasn't using a torque meter,a set square and a spirit level to ensure the PC was put together with with precision.
If a cable burns becasue it's not been put in just right that's not the users fault it's a problem with the design.
>If a cable burns becasue it's not been put in just right that's not the users fault it's a problem with the design.
And how many times has a user said their laptop isn't charging, only for the support person to discover, its not plugged in correctly to the wall socket or (more frequently) the mains cables isn't fully seated in the adaptors socket. Obviously, in these cases the fuse (in the plug) or the RCD circuit breaker tend to prove their worth.
The real concern has to be whether the cable actually burns ie. bursts into flames or not.
So when Americans accidentally zap themselves to death by touching the pins while inserting a mains plug, we should... sue the govmt? Disclaimer: I live in the UK where this can't happen, but it doesn't make the US system inherently bad. You overstress a heavy-duty power rail in any system, it's gonna cause issues, and until these keyed plastic molex-style plugs remain standard in PCs and while GPU power demand spirals ever upward I do't see how you can stop stupidity. Yes I said it.
This is a bad example, because you can sue anyone for anything in the US. Yes they could definitely sue the government for that (despite that not being a thing that actually happens because you don't hold US plugs by the contacts). People regularly slip on ice and sue the property owner successfully. It's definitely a thing to sue someone else for your own stupidity.
The law firm is probably from the same school of thought that considered a takeaway that supplied a coffee cup containing a hot coffee - the customer ordered, to be negligent for not labelling the cup "caution: contents hot". Ie. it made no difference to the way customers handled their cups of hot drinks, it just made it explicit the liability for the correct handling of the cup was the customer's...
...is actually quite interesting - you can watch videos about it on YouTube. Basically the coffee was deliberately kept at near boiling point (despite repeated warnings). The person who spilt it on herself needed multiple rounds of surgery...
It turns out that there is a legal limit to how hot you can serve beverages (about 60 centigrade or so) which is why they won compensation.
"there is a legal limit to how hot you can serve beverages (about 60 centigrade or so)"
So that's why it's always lukewarm-at-best by the time it actually turns up.
There ought to be an "I know it's hot and I won't shove it by my genitals" option to choose to get something that's, you know, hot.
Rubbish! Of course you can drink 60C hot water, that or I am some sort of freak as I drink coffee at 60C all the time.
I have a Dolce Gusto coffee machine ( about £40 off Amazon ) , I never use it for pods anymore simply as a low energy hot drink maker ( only 1000W for 30 secs ) for making instant coffee. I measured the heat on the water directly as it came out of the machine and it's at 63C-68C and I drink that out of the cup immediately as soon as the cup is full. It's it's the primary reason I keep the machine simply because I can drink it without having to wait for my coffee to cool.
The reason you can't use 60C water to make tea, tea leaf must be steep in very hot water, very quickly and hence why the Dolce machines can't be used to make cups of tea as the cheap ones are simply not hot enough.
I've written about the hot coffee lawsuit before, maccy Ds served her boiling hot coffee in a cup that was structurally not sound, she had terrible burns as a consequence and wanted Ronald to pay for some of her medical bills.
The big clown refused, she sued, and the jury were so appalled at the hamburglar's negligence and attempts to get away with it that they awarded a settlement much higher than she had asked for.
Two things happened as a consequence of this case, McDonalds serve their coffee at a lower temperature and use more sturdy cups and caps that are very unlikely to fall apart, and secondly, a PR firm was engaged to slander this poor woman with Isn't it ridiculous, woman burns herself with coffee, and WE'RE liable?!
Sadly, the slander really took effect.
Look at any part of the electrical code, and it will spec out rules to build a system that has enough overhead for safety to reliably operate without catastrophic failures in the real world. A design that is barely able to handle the power of a launch GPU that isn't overclocked yet, but can be is already broken. One that allows for fiddly connections quite easily, when the result is an electrical fire isn't suitable for purpose. 12 pins are not better than one, and as others have pointed out, if they were going to make a whole new connector, they could have had a pair of fat wires and four thin ones for the temp/power signals. With that if the contacts aren't fully seated your card won't start unless you plug it in all the way, instead the build a new connector that would try to pull too much power on the remaining pins if some weren't fully seated.
They chose a bad design, ignored early reports of problems, then tried to cover up the problem by blaming the victims of their faulty manufacturing. As a result they are being sued, will either settle or lose, and will probably have to recall all of the cards, cables and power supplies in this series. It's the galaxy note 7 of video cards. They can limit the brand damage if they handle this pro-actively, but it looks like they are going to try to sell them through the holiday season. Once enough of these are in the field, there will end up being a fire or accident in some kids room and the fire department will point to a slagged NVIDIA card on camera. That is the trajectory they seem to be choosing.
It's sad because in the long term the cost of replacement will end up a rounding error, but they will end up paying it regardless. If they handle this in a stand up way gamers will forget it by the new year. If they drag it out, it will be brand toxic, they will lose in court and it will cost more for something they will have to do regardless. Short term denial that this will all go away will be an expensive long term mistake. Part of NVIDIA's ongoing success is the boost they get from the network effect of developers optimizing for their platform, and that could slip along with their brands halo if the gaming community turns on them. AMD has a good recent track record for capitalizing on their competitors slip ups, and they aren't that far behind. Why give them the chance to leapfrog the market? Pick a better power link, and re-manufacture the old cards and dump them on the budget segment to recoup some of the loss.
Is that even if they are initially seated well, send a load or high heat through them and they can move if not positively connected. This can then cause arcing, higher heat, rinse and repeat until melty goodness or worse.
AC current is even more fun since it literally vibrates. This can cause cables to become loose, again arcing ensues and melty goodness. Really annoying when it causes your otherwise perfectly fine heating timer to get melty. Thankfully not burnie too (good design and materials) otherwise electrical fire next to water heater is asking for a visit from the local fire brigade and awkward questions from building owners.
NVIDIA knew when they designed it that it was going to be a tall card and secondly given the size of most computer cases that there would be a much smaller gap between the top of the card and the side panel requiring that the cables somehow be bent out of the way.
You wouldn't need to be an Einstein to figure out that the best solution from the outset would have been something like a right angled connector like we also do have for SATA drives as they also work very well. There's also no room for wires and connectors inside the plug to move past a 90 degree bend either so once it's plugged in you would know each individual connector would make full contact.
I guess they would've had to have two of them for each card such as an upward and downward connector depending on how the power cables are routed in everyone's cases and I presume further that saving a buck of two just providing a single substandard straight connector on cards that sell for thousands anyway somehow made good financial sense.
Look, I don't want any toast, and he doesn't want any toast. In fact, no-one around here wants any toast. Not now, not ever. No toast.
Or muffins. We don't like muffins around here. We want no muffins. No toast, no teacakes, no buns, baps, baguettes or bagels. No croissants, no crumpets, no pancakes, no potato cakes, and no hot cross buns. And definitely no smegging flapjacks!
NVidia is now aware of the problem and will obviously take steps to resolve the tech.
Some people will have been affected by the issue and it is very likely they will have to pay to fix the problems. NVidia will most likely attempt to address the financial issues and may even have to issue a recall on the damaged boards.
This lawsuit however will place NVidia on the offensive which will either delay financial reparations to the consumers and possibly even delay a recall as their legal team will most likely advise them that it will cost a great deal more to admit fault.
In the end, the affected consumers will probably receive checks for $50 if they live in the USA and the lawyers will make millions.
This is an excellent example of how class action suits can seriously backfire. The consumers will be hurt and NVidia will end up paying less than if than if they actually fix the problem.
"...considering that electric vehicle maker Tesla had to recall nearly 130,000 of its cars earlier this year due to overheating Ryzen processors used in the infotainment system."
Is that AMD's fault? Because there's no widespread reports of Ryzens overheating - only in Teslas.
Which makes me wonder whether it boils down to the chief Twit and the kind of engineers who'll put up with his management style... ("The bad ones left. I have all the good ones.")
According to the linked-to article, the Tesla is reducing the amount of cooling given to the AMD processor - guess what happens if you don't provide adequate cooling to a big CPU!
The article also says that other Tesla models use an Intel device that isn't overheating - because it has a lower TDP. Which doesn't mean that the Intel device is better, it means that some engineer(s) decided to bung in a different part without comparing the specs and without knowing how close to its limit the Tesla's cooling system must be! The latter alone is not a comforting thought.
As already pointed out in comments here, having failed to notice what could go wrong, they also failed to set up the AMD device to protect itself (by starting to throttle from a lower temperature before overheating causes more a dramatic response).
So, to answer the question "Is that AMD's fault?" - nope; pretty sure they publish the TDP of their devices, so it is then up to Tesla engineers to read *all* the relevant spec sheets and put two and two together.
Unless, of course, the Tesla internal spec sheets don't accurately describe the limits of the cooling system...
Highly unlikely: Those 5 Amps of USB-C are spread across four power connectors + four ground connectors. Not like the one-pin = 10 A of that ATX3.0 plug. But you miss something: Up to 48 V with 5 amps are allowed. But I doubt that it will be a problem since the negotiation must be complete before switching to higher voltage and amps, including checking whether all pins have contact.
It's not necessarily less important, given that they ran safety systems through that part (they had to recall cars over that system failing because all alert noises only sounded if that system was online). If they fixed that, it would still not be great for them to cut the power to something that is used to display things like views from cameras the driver might be using. The driver should be able to deal with the loss, but that doesn't make them expendable except when absolutely necessary. If the cooling isn't sufficient to prevent the batteries or the control system from overheating, then the only viable solution is to increase the available cooling or decrease the heat produced to a level the existing cooling can handle.
Just a bit of perspective; I have an AMD Ryzen. I had the case sitting on the floor where it'd been merrily collecting dust since 2017. On the most recent steam survey that I did several months ago, I realised from the results that actually, it was underperforming by quite a lot, even compared to other Ryzens of the same vintage.
It turned out that the inside of the case was liberally covered with dust, and instead of just tripping the thermal cutoff and instantly powering off in the way that Intel CPU's do [or at least used to when I owned them] the Ryzen just kept throttling the performance down to maintain a sensible operating temperature at the expense of performance. Given the sheer performance of modern equipment it was only actually noticeable from comparative benchmarks.
After cleaning the case the performance leaped back to where it should have been, so I suspect that you'd probably find that in a Tesla the interface just started getting slow and DVD videos being played started juddering.
That is, from my point of view, a Tesla bug. The CPUs can limit themself to any given temperature configured before going to hard-brake more 'cause they've reached 95°C or 110°C.
I use a temperature limit of 85°C set in overclocking setup to keep the CPU a bit cooler, effectively 5°C cooler than the allowed 90°C. So the method is there.
Monitoring whether all pins are seated well enough should be there, even though it might impossible in the way it is currently used. Would require a re-design of the board. And a warning from the driver. Additionally a temperature monitoring near the plug, integrated on the card.
When comparing to the normal 8-pin: If they are not connected correctly I hear my PC beeping, with a Titan RTX. Why didn't NVidia implement that? Feature over safety.
The power supply detects low-voltage situations and reacts by increasing the power-supply voltage to get the correct value at card, allowing for cable drop. This is normal, it applies for other devices in the box, that stuff about 'measuring the wattage' is people who don't understand what they are talking about.
There really isn't much the card can do about it -- the whole point of the system is that it maintains a steady 12V at the card.
The problem is, that doing that with a faulty high-power cable and connector system just increases the power into the fault.
There should be more monitoring at the Power Supply. The power supply should cut out when the cable drop is high on a high-current connection. The world may move that way eventually.
In any case, high current devices like this probably should use expensive, inconvenient, high-current connectors, like those used on solar panels -- the solution they've attempted hasn't worked.
It is possible to detect resistive faults such as a cable not completely plugged in or a bad connection from FOD (both seen as problems with the new connector). There is no need to give any warnings at the driver level, simply turn the card off (or don't even start it) when a fault is detected.
The first thing you (should) do when troubleshooting a graphics card problem, especially a newly installed one as would be the case for almost of these faults, is check and reseat power connectors.
Of course the hardware to detect those faults isn't present on the 4090 board, so we're back in redesign territory.
"AMD best be careful tossing stones in silicon houses, considering that electric vehicle maker Tesla had to recall nearly 130,000 of its cars earlier this year due to overheating Ryzen processors used in the infotainment system."
Because AMD designed the cooling system in the Tesla?
It was yet another Tesla screwup. The cooling system in the infotainment center was inappropriately turned off during "preconditioning" of the battery in preparation for, or during, a fast charge. That has nothing at all to do with AMD or any AMD design-- that's all Tesla. By contrast, this overheating connector is an nVidia design through-and-through.
There have already been issues with USB-C connectors used for power and expecting 5V, and USB-C chargers which do not negotiate correctly and assume max power rather than minimum. Examples include Raspberry Pi which is a physical USB-C power connector but only expects 5.1V@up to 4A ... an incorrectly or non negotiating power supply which has a USB-C connector (because it's the law in the EU) but supplies 12V fits fine and blows its sock off. Multiple voltages on a single power supply ... great idea EU ...
That is a know bug of the raspberry pie 4 series production run, implementation flaw. Fixed long ago.
Admittedly, it's probably running close to idle if it's just handling you typing a message. If you ran CPU and GPU-intensive tasks, that 100 W figure will get a lot higher. I'm not sure what your CPU's limit is, but the GTX750 can consume up to 55 W on its own if you stress it. I agree with you about the power consumption being crazy, and most users will not need anything close to the performance that this GPU provides. Your card has 512 shaders, whereas the one mentioned in this article has 16384 with another variant that goes up to 18176. You have to do a lot to justify something that performant. I'm not a gamer, but even the ones I know don't appear to need that much performance enough to justify the power consumption and cost.
It's good to know that a correctly installed 4090, with the connector plugged in carefully, is unlikely to have this issue.
However, does this sort of thing routinely happen on other video cards? Like the Nvidia 3000 series, or AMD's cards? If not, then clearly plugging in the connector just right has suddenly become a lot more critical than it used to be.
So Nvidia ought to be pulling the 4090 with the 12-pin connector, and replacing it with one with a different arrangement, better able to handle its demand for power. It might look to IBM for examples of how a serious company handles an "Oops" moment.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
