This was solved a decade ago in the enterprise world with an attachable bracket on the end of the card that fits into a slot at a fixed width so it is supported in the vertical axis by both the standard rear case slot and the new support on the forward side.
Nvidia had provisions on their Quadro's to bolt this bracket that all HP Z-machines used and I believe some Dell workstations as well.
The fact that these new cards are even heavier than a K5000 and yet there are not supplied attachment points for such additional support in some way sucks. It comes down to the fact that self-build PC case manufacturers aren't working together and with card suppliers to have a standard- nay, ANY provision for additional support.
Unlike in the enterprise world where the entire system is engineered at an extremely high level with communication from the vendors with the OEM (like a HP Z 800 machine).
That's it. It's not like HP had some magical patent on "hey just have a little extra bit of metal slide into a slot ad a fixed width inline with the PCIe slot". In fact, they made it extra-wide and attached to the hard drive bays in the Z800/820/840s as if they were anticipating modern cards to be bigger than the Keplers/Pascal cards at the time.
---
That said, the level of cracking shown here on the FR4 of these 4090s can only be the result of some pretty significant impact loads. These things were installed unsupported in a case and got dropped off the back of a truck, for sure. Or they were actually pulled out of a dumpster by an employee at a facility that had literally been doing fatigue vibe table tests on machines for an OEM...
It’s not even just the enterprise world. It’s in the PCI spec. It’s just that no one uses it.
There is a formal definition for retainer brackets. No one manufactures cases that allows them to be used though. And no card manufacturers integrate them.
I have a Gigabyte RTX 4080, it’s the same size and weight as a 4090. The card is over 340mm long and tall enough to occupy more than three PCI-e slots in my case. It weighs more than many laptops, and I had to add a physical reinforcement bracket to keep it from sagging and straining against the PCI-e slot it’s plugged into.
Looking at the article it’s fairly obvious the damaged cards came from system builder(s) who thought it would be sufficient to use two PCI-e bracket thumbscrews to secure the cards before throwing the whole PC in a shipping box and letting God sort it out.
Sure GPUs are becoming too large, but issues like this could be solved with a little care during assembly. Many high-end GPUs actually come with the necessary support brackets, which should be a hint.
I have to agree, I feel like the idea that maybe some extra support is necessary really started to hit mainstream around the 20X line of cards.
I have a 4090 in my machine, I straight up refused to install it and sat on it for a few days until I had a support bracket in place.
I think we could argue that at this point the cards should likely come with one in the box, but I think like you said a little bit of care when you are building your system.
I don't think this is a flaw with the card and really just a flaw with how we build computers and this connector. At best we could say the card shouldn't exist if it has to be this large but I don't think that is the solution.
Honestly, I'd like to see some science on the bracket design.
My card (an ASRock 6900XT, 332mm, 1.78kg per reviews) came with a bracket that's a modestly beefy-looking piece of steel, but it just bolts into two or three more thumbscrew sockets, and the actual contact-point with the card is a glued on rubber bumper that's like 2mm x 15mm and a rubber bumper, so there's opportunity for it to bend or not align with the card well, and the bumper gets unglued, and it blocks all your slots.
I ended up getting the type of bracket that's a rod with a little magnetic foot to anchor it to the bottom of the case, and then another foot that the card rests on. Again, this is dependent on how tightly the thumbscrew holds the top foot in place, but hopefully the downward force from the card is being projected straight down the bracket.
The right answer is probably to reintroduce flat-mount 'desktop style' cases. There was zero sag risk when I had the build in a CoolerMaster HAF-XB, but it was a bit shallow for a card that big - you couldn't get the card out without disassembling everything and lifting at odd angles while applying enough force to rip the clip off the end of the PCI-E slot.
Alternatively, if we could get some collaboration in the industry, they could agree on a standard "max length". I note there are some documents suggesting the maximum length for an ISA or PCI card used to be ~338mm. In that scenario, the cases can be designed with an explicit assumption there will be a 338mm card installed, and there are little support rails at the front to engage a card of about 338mm long. These could help bear the load and ensure the card remained aligned.
Ancient PCs had this-- I can recall opening an old Compaq 286 and seeing it stuffed full of "full length" cards neatly fitting into rails in the case. Back then, the cases were desktop and the cards were light, so the rails were more for management and neatness, but there's no reason they couldn't carry weight in a modern revision.
I think I've also seen it with some weird OEM builds more recently-- they'll add a metal or plastic shim to the front of a long card so it matches up with rails in the one case it was designed to fit into.
Gigabyte has a great support bracket design that screws in to two ATX holes on the motherboard and holds the card up from underneath[1]. It's one of the best designs I've seen because it bolts the weight across two planes on the case, but requires the PCI-e slots to be in a standard configuration (not a given on every board - it fit on my Z790 board but not an X670e board that I tried)
I think flat desktop cases might not be a bad answer, especially with the useless stands that come on most monitors these days with no height adjustment. Typically I need to raise a monitor several inches to not have to tilt it up and stare down at it, which is terrible posture. Going back to the old style (CRT sitting on top of a beefy steel desktop box) wouldn't be so bad.
Otherwise, even without an industry standardized max-length, enthusiast case manufacturers could solve this by providing their own brackets that span the width of the case and screw onto matching holes across from the PCI-e mounts. They could even put RGB lights on them to make them look less ugly for people with glass side panels.
Nowadays I have a RTX 3090ti, which is even longer than a 4090.
This solution isn't for everyone, but I made my pc case out of makerbeam rails, hanging the GPUs vertically, attached to the motherboard with PCIe x16 riser cables.
Some motherboards and pci slots have various methods of "locking" a card in place. Seems like this may have been forgotten and force was used to remove the card. I don't see how the weight of the card could crack the PCB, they are very durable.
Nvidia had provisions on their Quadro's to bolt this bracket that all HP Z-machines used and I believe some Dell workstations as well.
The fact that these new cards are even heavier than a K5000 and yet there are not supplied attachment points for such additional support in some way sucks. It comes down to the fact that self-build PC case manufacturers aren't working together and with card suppliers to have a standard- nay, ANY provision for additional support.
Unlike in the enterprise world where the entire system is engineered at an extremely high level with communication from the vendors with the OEM (like a HP Z 800 machine).
That's it. It's not like HP had some magical patent on "hey just have a little extra bit of metal slide into a slot ad a fixed width inline with the PCIe slot". In fact, they made it extra-wide and attached to the hard drive bays in the Z800/820/840s as if they were anticipating modern cards to be bigger than the Keplers/Pascal cards at the time.
---
That said, the level of cracking shown here on the FR4 of these 4090s can only be the result of some pretty significant impact loads. These things were installed unsupported in a case and got dropped off the back of a truck, for sure. Or they were actually pulled out of a dumpster by an employee at a facility that had literally been doing fatigue vibe table tests on machines for an OEM...