It’s not really an understatement to say that over the years videocards (GPUs) — much like CPU coolers — have become rather chonky. Unfortunately, the PCIe slots they plug into were never designed with multi-kilogram cards in mind. All this extra weight is of course happily affected by gravity.
The problem has gotten to the point that the ASUS ROG Astral RTX 5090 card added a Bosch Sensortec BMI323 inertial measurement unit (IMU) to provide an accelerometer and angular rate (gyroscope) measurements, as reported by [Uniko’s Hardware] (in Chinese, see English [Videocardz] article).
There are so-called anti-sag brackets that provide structural support to the top of the GPU where it isn’t normally secured. But since this card weighs in at over 6 pounds (3 kilograms) for the air cooled model, it appears the bracket wasn’t enough, and active monitoring was necessary.
The software allows you to set a sag angle at which you receive a notification, which would presumably either allow you to turn off the system and readjust the GPU, or be forewarned when it is about to rip itself loose from the PCIe slot and crash to the bottom of the case.
Bonkers that we have gotten to the stage that sensor makes any sense at all, though that is a great way to differentiate your product and could well become the default wanted by anybody that still LAN parties and system integrators that have to ship these monsters pre-built…
Shouldn’t this be “ASUS GPU Uses Accelerometer To Warn For Sagging Cards”? By measuring the static acceleration due to gravity, you can find out the angle the device is tilted at with respect to the earth. A gyroscope measures angular velocity (rate of rotation).
Looking up the chip it’s a 6DOF IMU. That said you are right that using the gyroscope for any of this seems silly. Except to maybe filter out “alert someone just smacked the case”
I would call it a design flaw if the card, in its standard-conforming mount of the slot(s) and PCI interface, can’t support its own weight in a vertical case. Especially since that’s pretty much the default configuration nowadays. Some case manufacturers have attempted to implement a support structure, but it’s hard to design for when the card lengths are effectively random. Not to mention damaging the PCI socket or even the motherboard substrate itself.
It’s really not that hard to design for because there are defined, standard form factors for PCI-E cards so case manufacturers can build products to suit and provide the support rails for cards to slide into.
Obviously if card manufacturers don’t build cards to the standard “full length” dimensions then it really shouldn’t be too much to ask that they design and supply a bracket that bridges the difference so your super expensive GPU isn’t flexing the BGA chips off the board?
IBM Microchannel was designed with a support structure made in cast alloy, and Microchannel board were made with a blue plastic stub that slid in the support structure. Shorter board had a longer stub.
Even the original IBM AT PC had support slots for supporting the longer ISA expansion boards.
Pretty much every name brand PC I’ve ever encountered (and I have seen a LOT, I’ve been ‘doing IT’ for near 40 years now) has had those support slots, all the way from original IBM PC to the current 2024 HP Z Workstation and the 2023 Optiplex under my desk.
The RTX5000 in the Z Workstation has a decent bracket to engage with those support slots too.
This only seems to be an issue with aftermarket cases and cost cutting GPU manufacturers.
those standards were meant for single slot cards without a heat sink. gpu cooling requirements keep demanding bigger heat sinks. but i think we got to the point where the mounting system is no longer adequate. both ends need to be supported by the chassis.
vanity slots kind of solve the problem in a way, provided you need to use a riser cable. i really dont like those, mostly because they are meant to show off your gpu but at least its not supporting it on the shortest axis.
horizontally mounting the card directly to the mobo tray (below the mobo) with tapped holes and standoffs might be a good idea. tapped holes and standoffs are easier to add to a case standards than double brackets or rails. cases usually have more screw holes than you can use and adding a few more and a few extra standoffs to the parts baggie is possibly the easiest solution for case manufacturers. probibly more difficult on the gpu side as the fans would get in the way. i figure the pci connection can be either a u-bracket or riser cable. later on you might mobos design their pcie slots specifically for this arrangement. it also clears the board so you can get to your m.2 drives without taking out the gpu.
the design of some of the 50xx cards the pcie slot is on a daughter board anyway, and so theoretically you can put it in any orientation you want. still waiting for an sff build with a monoblock combined mobo+cpu+gpu cooler with the boards hanging off the block rather than the other way around.
You’ll need an adjustable tower case so the videocard will be horizontal. We’ll call it PISA.
You know, there once was a time when many PC cases were designed to lie flat on the table beneath the monitor. They called it “desktop” PC case.
I was there when they were invented. Also I was present when the humor sense was invented ;)
Or just use a vertical mount kit/case. Much better.
Silverstone Raven here. Solves the weight problem easily.
That is insane.
Is this 27 days late?
This is kinda hilarious, a 3 kilo graphic card. Literal newborn infants weigh less
It’s about the weight of a lot of rifles.
They’d be using the accelerometer not the gyro of the imu to measure the angle of the card with respect to gravity. The gyro is only good for quick relative changes in angle.
Time to flip form factors:
Put the GPU on the ATX board and add the CPU via PCIe-slot…
Or better yet, make two ATX boards, one for CPU and one for GPU, that can be joined edge-to-edge. Much more space for giant heatsink towers then.
Just make a slot in the GPU for the CPU and a larger slot for the PSU…
Who need motherboard anyway
the coolers weigh more than the cards, so i think we should mount them by their heat sinks, and let the boards hang off of them.
Back in the day, we had hard drives on cards, and the way this was solved there was a couple small pieces of channel aluminum that ran the full length of the card. How have we gotten to the point where we choose to detect mechanical failure rather than prevent it with reinforcement?
Well maybe this card hit a critical point in the “mass vs. additional support” graph where adding support isn’t feasible anymore because the additional weight would require even more support which adds more weight ….
:-)
Hanging an accelerometer off SMBUS and paying a few grand NRE for the software is far cheaper than machining a new support structure. That doesn’t make it the right solution, but margins even on high end GPUs are razor thin for the manufacturers and most consumers will end up buying on performance and completely ignoring reliability fixes. Note how many people still bought 4090s even though their new DC in connector design was known to spontaneously combust.
To be fair, what hard cards had in mechanical stabilization they lost out on in reliability and utility. They were a niche solution for a problem that all but disappeared as soon as 3.5″ IDE hit the market, which was only a couple of years after their release.
Indeed, my first HD was a Plus Hardcard XL 105 and had lots of black aluminium everywhere to support the drive.
Only got rid of it a few years ago, didn’t have a machine with an ISA slot to put it in.
Hah, that’s exactly the first hardcard I had, too, and what I was seeing in my mind’s eye
When all you know is EDA software, everything looks like a nail.
Don’t forget that pre built machines must survive shipping with these cards installed! A lot can happen in transit
Even for a custom build, it seems the damage would be most likely to occur when moving the machine, rather than when it is still and being monitored by the IMU.
If I remember correctly, even my Quadro FX 4600 came with a metal support bracket at the front side. It would slide in one of those plastic rails at the front of my computer case while inserting the card.
This same bracket was also present on some big SCSI adapters and other cards that would reach to almost the front of the case. Probably something server and workstation related.
I’m not sure if there are still any computer cases with these brackets, but wouldn’t it make sense to re-use something proven, that has been used for years? It was a really simple solution with the only downside that sometimes the HDD cage would be blocking the front of the case.
really a comic problem from my perspective…easy for me to laugh at, because i don’t play modern high-end videogames or run LLMs at home. pretty sure doom3 runs decently well on “Intel UHD” integrated GPU right?
but the mechanical problem reminds me of the old full-length ISA cards. there’s a little plastic slot in the very front of the case, which will grip the end of the card if you have one that long. i had exactly one card that was that long…i think it was pre-hercules MDA board? you know, 80×25 text with about 4kB of memory. monochrome green or orange monitor.
i’m glad to be completely ignorant of how that has evolved in the modern era. personally, just installing a modern pcie or M.2 card is a voyage of discovery each time. x1 vs x16, M/B-keying, is this M2 slot even mini-PCIe or something else? who knows. everything’s easier away from the bleeding edge
I dont think I have noticed slots in the front of cases for a while now unless its some workstation / server setup
even 20 years ago they were blocked off by hard drive cages going from top to bottom, and todays gamer cases have to cram in a 360 x40 rad and thick ass fans and all that tubing :)
I wonder why GPU makers don’t implement telescoping rails that project out from the end of the card, which can be attached to a bracket that uses screws or 3M’s heavy-duty double-sided mounting tape to attach to the side of the case. Sure, there may be fans or other obstacles in the way, but you can just sell optional brackets to work around them. Or provide files for users to 3D print them themselves.
definately a hack.
i think its about time the atx standard gets ammended to come up with a proper structural mounting methods for gpus that weigh more than a cat. 2-3 screws and an armored slot do not a mounting system make. you need a 4-point mounting system for something that big. at least have brackets at both ends. direct to chassis cpu cooler mounting would also be a good idea given the size of some of the better air coolers.
The original XKCD phone would scream when falling https://xkcd.com/1363/
— and somebody implemented this: https://apt.steverolfe.com/depiction/freefall.html