We Need New Motherboards Before GPUs Collapse Under Their Own Gravityerikmcclure.com
Taniwha 373 days ago [-]
This is not a new problem, back in the late 80s I worked for a Mac graphics card developer ... We made some of the first 24bit accelerated graphics cards.

Our first was just an existing card with a small daughter card with pals an sram on it, it was so easy that we got our own logos put on many of the chips to put the competition off the scent, we designed that one in days and got it to market in weeks.

We immediately started on 2 more designs. The next was all FPGA, it was as big a nubus card as one could build, it pulled too much power, and tilted under it's own weight out of the bus socket (Mac's didn't use screws to hold cards in place, that happened when you closed the chassis). We got it out the door about the point that the competition beat the first board's performance.

The final card was built with custom silicon, designed backwards from "how fast can we possibly make the vrams go if we use all the tricks?", In this case we essentially bet the company on whether a new ~200 pin plastic packaging technology was viable. This design really soaked the competition.

In those days big monitors didn't work on just any card so if you owned the high end graphics card biz you owned the high end monitor biz too ... The 3 card play above was worth more than $120m

smilespray 373 days ago [-]
I may well have used those cards for work! Radius, RasterOps or SuperMac?
Taniwha 373 days ago [-]
SuperMac - the first one was the "SuperMac Spectrum/24 series III", I can't remember the name of the big FPGA thing

I built the accelerator on the last project (that 200 pin chip - look for a chip SQD-01 aka 'SQuiD') that was the basis of a whole family of subsequent cards (starting with the Thunder cards)

smilespray 373 days ago [-]
I loved those cards. My daily driver was a fully stacked IIfx with a SuperMac card. I remember telling my friends my display was more than a thousand pixels (1152 or something) wide.
Taniwha 373 days ago [-]
yup 1152x870 - essentially the largest number that is roughly a 4:3 aspect ratio and less than 1 megapixel (2^20 VRAM was so very expensive) (and meets a bunch of other limitations like video dot clocks, VRAM clock rates, must be a multiple of the width of the VRAM banks etc)
372 days ago [-]
im_down_w_otp 372 days ago [-]
I have a SuperMac Thunder/24 w/ DSP upgrade currently in a Quadra 840AV in my little home office museum here. I even have the original boxes, software, etc. :-)

I'm a fan of your work! I'd love to chat sometime.

Taniwha 372 days ago [-]
I'm not sure how useful I could be (it was more than 30 years ago) - remember that while I designed the accelerator these boards were made by a team of smart people, wasn't just me
tambourine_man 372 days ago [-]
I’d love to know more about that, if you blog or podcast, please share us a link.

I never owned one of those cards but worked on a Mac that did on a summer job as a teen.

rwmj 373 days ago [-]
I was going to guess TARGA, but maybe those cards were PC only?
dylan604 372 days ago [-]
I remember back then searching ebay for a used Targa, but being confused by why all of the result thumbnails were Porsches.
analognoise 373 days ago [-]
This might be a weird question, but do you have any rundowns of the architecture handy?

I love Retrocomputing and it sounds like an awesome thing to look at.

Taniwha 373 days ago [-]
Nothing in particular - VRAM was the core of all the cards thise days, the serial port shift register (inside the VRAM) fed muxes that in turn fed luts and out to the display.

The VRAM array was 2 sets of 96 bits wide - the accelerator was in part a 1:4 24:96 mux that drove the VRAM array - that's how the dumb frame buffer worked.

I'd spent a year before all this writing code to grab stats figure out where the mac graphics code spent it's time, initially we'd expected to optimise drawing vectors (ie lines - autocad was driving the parallel PC graphics market at the time) - my big discovery was that for the sorts of workloads that our customers were using 99% consisted of just 4 operations:

- solid fills (>90%) - patterned fills - copying within the FB (ie scrolling) - writing text (quite low by important for the DTP market)

I wrote the spec for the first 2 cards (to do the first 3 operations) and the went off to design the 3rd chip (modelled in C, turned into an inch thick spec describing the state machines and data paths and then passed to someone else to make gates)

I added a hack on top of the copy primitive to accelerate text hiliting.

So that accelerator chip also contained a state machine did 4 simple things as fast as it possibly could (solid fills used magic modes in the VRAMS, it could fill at 1.5Gbytes/sec - faster than the high end 3D hardware at the time)

In the end the text fill was useful, but not that much faster (it turned 1-bit font images into 24-bit pixels on the fly), all that cool hardware was pissed away by the way the underlying apps used the font rendering subsystem (page maker invalidated the font cache every line, quark did an N-squared thing, etc)

userbinator 372 days ago [-]
my big discovery was that for the sorts of workloads that our customers were using 99% consisted of just 4 operations

Those are the same operations that most if not all of even the very cheapest "2D-only" integrated graphics are capable of, starting from the early 90s up to today.

Taniwha 372 days ago [-]
well this was 1989 ..... as I said prior to this state of the are was fast vectors for CAD, and before any of that we didn't have many per-pixel framebuffers (memory was so expensive so you got stuff like CGA/etc or plain ttys)
ascar 373 days ago [-]
I'm surprised this isn't mentioned here more clearly: Some high end cases like the be quiet silent base I'm using have the option to mount the graphics card vertically, basically parallel to the mainboard, in a separate anchor slot. It needs an additional connector cable (~$20), but other than that is easy to setup, looks better with a look-in case (the illuminated fans face the glass side) and the weight pulls on a special anchor point just for that with no electronics involved. Plus the card itself is more sturdy in that orientation and there is no issues with bending through its own weight. It might even be beneficial ventilation-wise as the graphics card no longer causes a horizontal divide (basically creating separate ventilation-zones on the top and bottom of the card).

Yes, the cable will add approximately ~0.3ns of additional latency due to the added 10cm of distance.

This is how it looks like:



mastax 373 days ago [-]
These 4-slot cards won't fit in most vertical mounts, and if they do the fans might be pressed up against the side panel. But that's definitely a good route to go, now that case manufacturers know what they're dealing with.
ant6n 373 days ago [-]
Or simply use a case which has the ports at the top, so both motherboard and graphics card are vertically oriented, but still plugged in normally. Case fans are at the bottom and push air up out the back slots at the top.

Example: Silver Stone Fortress case https://bjorn3d.com/2010/05/silverstone-fortress-series-ft02...

Hamuko 373 days ago [-]
I like my mini-ITX case (Phanteks Shift Air 2) because it hangs the GPU from the top and it's sitting behind a mesh so it still has ample access to fresh air.


tcmb 373 days ago [-]
I'm getting a Cloudflare error page on that link: 'The owner of this website does not allow hotlinking to that resource'
Hamuko 373 days ago [-]
Huh, didn't realise it spits out that kind of an error. I guess Cloudflare whitelists your IP address if you visit it once since that links still works for me without visiting the original URL?

Anyways, you can see the photo here: https://www.kitguru.net/components/leo-waldock/phanteks-evol...

tecleandor 373 days ago [-]
Yep, it's probably the deep linking protection. It probably doesn't like the HN referer
patates 373 days ago [-]
I could open the link in a private/incognito window.
dotancohen 373 days ago [-]
Same here
justsomehnguy 373 days ago [-]
Wow, this is almost exactly as I wanted (except I wanted a horizontal mount akin to blade servers). Never seen it, thanks.
rkangel 373 days ago [-]
This is a helpful post because I have exactly that case and no idea what that slot was for!

It's a fantastic case.

adwf 372 days ago [-]
I was going to say much the same. Vertical mount is a game changer.

Whilst this example is a bit on the extreme side:


It does have some nice properties, no GFX card sag, any water cooling leaks can't drop on expensive components, power supply is back and away too.

Most important point is the cables out the top are covered by a shield so you don't have the ugliness.

dQw4w9WgXcQ 373 days ago [-]
How does latency impact rendering etc?
danielheath 373 days ago [-]
PCI latency has no impact on rendering, but your card/motherboard may fallback to PCIE3 instead of using PCIE4, in which case filling the cards entire VRAM from RAM could take an extra half-second (slowing the loading of new textures an infinitesimal amount).
mtsr 373 days ago [-]
Modern engines also do a lot of streaming data to VRAM, rather than loading everything up front. So it’s not necessarily a single delay and could very well impact framerates etc.
eropple 372 days ago [-]
Is there anything out now, or even on horizon, that's materially impacted at running at PCIe 3.0 rather than PCIe 4.0 rates? I can't think of much.
userbinator 373 days ago [-]
Every time I see the sizes of GPUs increase, I'm reminded of this from over 2 decades ago:


l-lousy 373 days ago [-]
The fact that even the “not to be sold in china” bit is so on point even today is kind of scary
hutzlibu 373 days ago [-]
Or it is relaxing to be reminded, that all those geopolitical tensions and struggles are basically nothing new, so should not be taken too seriously.
0x38B 373 days ago [-]
Haha: "Bitchin'fast!3D²⁰⁰⁰ supports neither OpenGL nor Direct3D, nor does it fit in any case known to man."

Almost like some giant graphics cards...

somat 373 days ago [-]
The sgi onyx reality engine "graphics card" was composed of like 5 motherboard sized pcb's.


bofh23 371 days ago [-]
I think it was 3 to 6 VME-sized boards. There was 1 Display Generator board, 1 Geometry Engine board (12 x Intel i860 processors in the RealityEngine2), and 1, 2, or 4 Raster Manager boards (for framebuffer and texture memory). I’m not sure if the maximum configuration of 6 boards could fit in the deskside chassis. That might have required the full rack chassis which had space for several sets of cards for multi-pipe systems. As for cooling, I know the full rack chassis had a huge blower. I’m not sure if the deskside models had blowers or more traditional axial fans.
drewm1980 373 days ago [-]
415 Bungholiomarks ought to be good enough for anyone.
greggsy 373 days ago [-]
This has to have been before every graphic card advertisement included a 3D rendering a scanty-clad heroine.
krasin 373 days ago [-]
Here is a video review of the BitchingFast 3D 2000: https://www.youtube.com/watch?v=ZmCEXbspWuI
sqrt_1 373 days ago [-]
The same Youtuber did a 4090 "review" for April fool day https://www.youtube.com/watch?v=0frNP0qzxQc
bl0b 373 days ago [-]
"256 MB of curiously high bandwidth"
deathanatos 373 days ago [-]
If you watch the linked YouTube video, one of the card's marketing materials — which clearly exactly 0 people read prior to publication — is quoted (screenshotted, in the video[1]) as,

> has a total length of 2704mm and a surface area of 905637mm²

We're going to need a bigger case.

[1]: https://youtu.be/mGARjRBJRX8?t=990

patates 373 days ago [-]
Looks like not only won't it fit to any case, it may even have some problems fitting in some smaller home offices as well. But hey, it doesn't need a support bracket!
tjr225 373 days ago [-]
This was a fun Time of humor: https://youtu.be/ZAtzN_ScKXY
altacc 373 days ago [-]
It exists! I looked at that and was sure it was a spoof until I saw the video of it.
serf 373 days ago [-]
yeah, but most of the weight on the new cards is just heat management stuff like heat pipes fans and heat-blocks/sinks.

That card is huge, but I bet it's a fair bit lighter than most of the gaming sector cards now.

neurostimulant 373 days ago [-]
> * ugly mutha heatshinks sold separately
dylan604 372 days ago [-]
Hmm, multiple GPU chips on the same board? Could this be where did Nvidia came up with the idea?
green-salt 373 days ago [-]
I am so glad I'm not the only one.
Lramseyer 373 days ago [-]
> Should we have GPU VRAM slots alongside CPU RAM slots? Is that even possible?

I chuckled a little at this because I used to wonder the same thing until I had to actually bring up a GDDR6 interface. Basically the reason GDDR6 is able to run so much faster is because we assume that everything is soldered down, and not socketed/slotted.

Back when I worked for a GPU company, I occasionally had conversations with co-workers about how ridiculous it was that we put a giant heavy heatsink CPU, and a low profile cooler on the GPU, which in today's day and age produces way more heat! I'm of the opinion that we make mini ATX shaped graphics cards so that you bolt them behind your motherboard (though you would need a different case that had standoffs in both directions.)

pvarangot 373 days ago [-]
I had the misfortune of bringing up firmware for an overengineered piece of crap with DDR5 and even though we had the proper measuring equipment instrumenting properly is barely possible. It's designed with reflections in mind and there's a loopback channel at a slower word rate that I think the controller uses to characterize the analog signal because it's like, barely digital.
Lramseyer 373 days ago [-]
Did Synopsys design the DDR5 controller by any chance?
pvarangot 368 days ago [-]
I don't remember but I think no, it was a Renenas chip? I think?
greggsy 373 days ago [-]
The NUC Extreme line is basically that, which IMO is a really good form factor.

Memory and nvme storage are dense enough to make anything larger obsolete for the practical needs of most users. Emphasis on ‘practical’, not strictly ‘affordable’.

The only advantages of larger form factors are the ability to add cheaper storage, additional PCI-e cards, and aftermarket cosmetic changes such as lighting. IMO, each of these needs represent specialist use cases.

Lramseyer 373 days ago [-]
Pretty much, but I'd change the dimensions of the graphics card to match mini ITX, and so that it could be bolted to the case. This provides two benefits: It can support a bigger and heavier heatsink and it also allows you to spread the VRMs around the chip and DRAMs for more consistent power flow.
YurgenJurgensen 373 days ago [-]
Okay, how about this: The PCI slot goes on the bottom of the mini-ITX motherboard, and extends out of the bottom of the case. The GPU is in its own enclosure, with a PCI edge connector on the top, and you stack one on top of the other.

I'd really like to find a way to daisy-chain them, but I know that's not how multi-gigabit interfaces work.

Raspberry Pi hats are cool. Why not big mini ITX hats? Yes, I just want to go back to the time of the Famicom Disk System, the Sega CD or the Satellaview.

mariuolo 373 days ago [-]
> The NUC Extreme line is basically that, which IMO is a really good form factor.

Missed that one. I still mourn BTX.

voxadam 373 days ago [-]
I wish a PC form factor like NLX[0] had become popular where the CPU and memory is on a board that's inserted into a backplane parallel to the add-in cards (similar to VME[1]). IIRC NLX was mostly intended for low-profile applications like corporate desktops and retail systems (point-of-sales) but it never caught on. I can see the edge connector and backplane potentially being an issue with old school parallel PCI (that's a lot of pins) but the serial nature of modern PCIe and the advent of PCIe switches would significantly reduce the number of signal lines.

[0] https://www.halolinux.us/hardware/images/2249_56_22.jpg

[1] https://en.wikipedia.org/wiki/VMEbus

nextaccountic 373 days ago [-]
> The NUC Extreme line is basically that, which IMO is a really good form factor.

Is there a case like this, but that is not part of a prebuild computer?

5d8767c68926 371 days ago [-]
I have been thinking about picking up this miniITX, Cooler Master NR200P[0]. Similar form factor that would be completely adequate for a daily driver and could accommodate a sizeable GPU if required. The problem is that the smaller builds are still niche so you have to pay a premium on an appropriate motherboard and power supply.

[0]: https://pcpartpicker.com/product/29drxr/

numpad0 373 days ago [-]
Yes, there is a standard called PICMG used in “portable/luggable computer” industry - oscilloscope-shaped PCs for field uses. Kontron and Advantech seems to be major suppliers there.
hinkley 373 days ago [-]
What if GPU makers made motherboards?
sgtnoodle 373 days ago [-]
I vaguely recall Nvidia motherboards were junk in the late 2000's.
7speter 373 days ago [-]
They made motherboard chipsets (nforce, iirc?), not motherboards, unless I missed something.

I think by the late 2000’s, though, their discreet gpus on laptops were problematic because they got so hot they detached from the motherboards or fried themselves. In a lot of cases, these systems shipped with these gpus and intel processors with integrated igpus.

This happened to a thinkpad t400 I had a while ago, the nvidia graphics stopped working and I had to disable it/enable the intel gpu in the bios (maybe even blind).

Iirc this snafu was what soured relations between apple and nvidia.

mschuster91 373 days ago [-]
> I think by the late 2000’s, though, their discreet gpus on laptops were problematic because they got so hot they detached from the motherboards or fried themselves.

That was indeed around ~2008-2010 - the issue was not that the chips fried themselves or got too hot. The issue was the switch to lead-free solder [1]... thermal cycling led to the BGA balls cracking apart as the lead-free solder formulation could not keep up with the mechanical stress. It hit the entire NVIDIA lineup at the time, it was just orders of magnitude more apparent in laptops, as these typically underwent a lot more and a lot more rapid thermal cycles than desktop GPUs.

> Iirc this snafu was what soured relations between apple and nvidia.

There have been multiple issues [2]. The above-mentioned "bumpgate", patent fights regarding the iPhone, and then finally that Apple likes to do a lot of the driver and firmware development for their stuff themselves - without that deep, holistic understanding on everything Apple would stand no chance at having the kind of power efficiency and freedom from nasty bugs that they have compared to most of the Windows laptops.

[1] https://semiaccurate.com/2009/08/21/nvidia-finally-understan...

[2] https://www.pcbuildersclub.com/2019/01/zwischen-apple-und-nv...

Arrath 372 days ago [-]
I recall my HP laptop with nvidia gpu getting painfully hot to the touch while raiding in WoW:TBC, probably could have cooked an egg on that sucker. It eventually gave up the ghost and I have to assume it was due to the heat.
Godel_unicode 373 days ago [-]
All the big ones do already.
hinkley 372 days ago [-]
Sorry, what I meant was, what if the GPU and the motherboard fused into one unit. So the CPU and main memory in but the card is fixed.

I guess the main problem with that we have this cycle where most people have a single graphics card with one or two monitors plugged in, and the power users have two of them to drive four screens, or to have 2 cards drive one monitor by daisy chaining them together.

But in the case of the latter two, it's usually a problem of a bottleneck between the card and the main bus, which of course if you were making the motherboard, you'd have a lot more say in how that is constructed.

staz 373 days ago [-]
What if we plug the motherboard in the GPU instead
neoberg 373 days ago [-]
What if gpu was the motherboard instead?
373 days ago [-]
qwerty456127 373 days ago [-]
Back in the days when I was a kid tower PCs were comparably rare and most of the PCs used the horizontal desktop design which essentially is the same like a tower but put on its side. People would often put the monitor on top of it to save the desk space (see the Windows 95-2000 "my computer" icon). Isn't it time for that to return so we wouldn't need “GPU Support Sticks”?

By the way, what actually dissatisfies me is the majority of mainboards having too few PCIex slots. Whenever I buy a PC I want a great extensible future-proof mainboard + very basic everything incl. a cheap graphics card so I can upgrade different parts the moments I feel like . Unfortunately such many-slot maininboards seem to all target the luxury gamer/miner segment and be many times more expensive than ordinary ones. I don't understand why some extra slots have to raise the cost up 10 times.

wongarsu 373 days ago [-]
> horizontal desktop design

That's be a pretty clever solution. I wonder if any case designers are brave enough to try to pitch that solution. Somewhat fits in the rising popularity of putting your tower on your desk.

> I don't understand why some extra slots have to raise the cost up 10 times

Part is market differentiation, but the other part is that you quickly run out of PCIe lanes to the CPU (especially with Intel, thanks to their market differentiation). If the mainboard manages to use no more PCIe lanes than the CPU supports, everything is simple. Once you add more or bigger PCIe slots, more M2 slots or more USB3 ports, the mainboard has to include a fairly expensive PCIe switch to turn few lanes into more.

qwerty456127 372 days ago [-]
I see. A very good point. Thank you. This reminds me about the explanation of why do USB-C docks have separate USB-2 ports: because every USB-3 lane has a separate USB-2 sub-lane and these are left vacant after using USB-3 lanes to connect an SD card reader and a NIC internally.

By the way, about brave horizontal desktop PC designs, note a neighbor comment by c0balt mentions one, to it seems very nice, I would buy one immediately if a traveling salesman would knock my office door offering it without the mental overhead of having to order it and wait.

hyperdimension 372 days ago [-]
Funnily enough, I happen to sell these horizontal ATX cases door-to-door! Only thing is, I'm a bit far from your area, and I can't stand travelling more than I need to. You'd help me a ton by helping me schedule my way to you, if you've got a few minutes.
dylan604 372 days ago [-]
> Somewhat fits in the rising popularity of putting your tower on your desk.

I would never want a PC on my desk again. The fan noise would be annoying as hell. Also, the form factor for modern multi-monitors would make them awkward to have sitting ontop of a box like that.

The modern designs are using silent systems like those from Apple with passive cooling, so you're not going to be using a GPU like those discussed here.

Kerrick 372 days ago [-]
My wife has a modern case with a horizontal motherboard: Thermaltake Core V21. It's not short enough to put a monitor on top of (because it uses a dual-layer design with a "basement" under the motherboard for the PSU and HDDs) rather than being wider, but it does at least solve the GPU leverage issue.
dawnbreez 372 days ago [-]
Another useful feature of the horizontal desktop design is that putting the monitor on the case brings it closer to a comfortable viewing level.
c0balt 373 days ago [-]
The problem with cases that are oriented horizontally is they're usually required to be in a slim form factor in comparison to "normal" ATX or even mini ATX cases. This poses a problem for providing adequate cooling and power delivery. In face of GPUs (NVIDIA/ AMD) becoming more power hungry/ heat producing this just doesn't work out for most people. There were some attempts in the past[0] however they don't appear to have gained mainstream adoption. On the other hand, systems with low power components could be accomdated in this form factor but they most likely wouldn't benefit much from it (or just employ a laptop with a dock).

[0]: https://www.youtube.com/watch?v=nRBJ2fysc64

372 days ago [-]
rkangel 373 days ago [-]
Horizontal form factor might solve the 'at rest' problem, but the moment you pick up the case to move it somewhere (even still while horizontal) the socket it going to be under load. We really need a solution that supports the card well enough that it's safe even in transit.
dtx1 373 days ago [-]
Because they usually have to use a different more expensive chipset and a lot of stuff uses PCIE these days that didnt before, for example NVME SSDs now take PCIE slots that were free to be used for different stuff before. And in the desktop market, most people just dont need more than a pcie slot for their boot drive and their GPU. Add to that that PCIE 4 and 5 are more expensive and require more expensive boards.
chickenimprint 372 days ago [-]
What exactly do you want to expand it with though? Sound cards are dead and everything besides accelerators is USB now.
373 days ago [-]
stormbrew 373 days ago [-]
The thing I don't get is why are we so damn stuck on towers as the default form factor? It's pretty much masochism to mount a sensitive piece of 8 billon layer circuit board vertically and then hang a bunch of blocks of solid heat conducting metal to it from the side, held on only by a soldered on piece of plastic.

Bring back proper desktop cases!

techdragon 373 days ago [-]
And while we’re at it, let’s bring back rack mount chassis. The rack mount market is appalling at the moment. If you don’t want the volume sales model imported to your region, your stuck getting overpriced models shipped at overpriced rates all the way from the increasingly limited number of manufacturers in china/India/etc, many of which are slowly sliding into overpriced industrial only markets. I bought one at full list price and shipping that had so little care put into it they didn’t even take the protective plastic wrap off the metal subcomponents before slamming them together! All because I wanted a generic 3U chassis to put together a compute node with low profile network cards. I looked at getting an empty chassis to just go open frame drill my own holes and it would have cost more to get a hollow sheet metal tube 3U high than the rack mount unit I bought… ridiculous!
stormbrew 373 days ago [-]
I had a similar experience trying to find shallow storage chassis that would fit in 2 or 3u. Seems like a no brainer to me to have just a rack mount array of 5.25 bays I can put hot swap cages in but nope.
techdragon 373 days ago [-]
Absolutely! It’s so bad that “custom” rack mounting is one of the reasons I got myself a drill press. Sheet and bar stock is trivial to order cut to simple dimensions for panels, corner bracing and making “rack ears”.

But it still comes back to the fact i shouldn’t need to make a mini metalworking shop in order to affordably mount hardware horizontally in my rack. Rack mount should be the ultimate in “big computer enclosure” but instead of cool things like having a 2/3/4U massive rack mount water cooling radiator with big quiet fans to silence everything in a computer, it’s getting harder and harder to buy any parts for a non-OEM computer that aren’t festooned in RGB and designed for transparent towers with all the structural issues highlighted in the original article.

ironmagma 373 days ago [-]
Right? Floor space is at an all-time premium with high housing costs. And most people I know have a stack of books/etc. to rise the monitor anyway.
tpmx 373 days ago [-]
Having a ~140 mm tall desktop case under the monitor doesn't work that well with today's pretty large monitors - it would result in uncomfortable viewing angles. Add to that the often very wide monitor stands. And increased cooling noise because the case comes closer to your ears.
mtsr 373 days ago [-]
Current monitors can be raised on their stand, maybe future monitors could be lowered in front of whatever they’re on top of?
Nevermark 373 days ago [-]
My 98 inch “monitor” is mounted on the wall

Great for demoing things, or looking at while I walk in think circles. Keeps my desk clear, and I don’t need reading glasses to see it

And yes, the bottom is mounted well below my desk’s surface height

naasking 372 days ago [-]
Mount the computer on the back of the screen.
tpmx 372 days ago [-]
You'd need counterweights in the front. :-)
naasking 372 days ago [-]
Or the case designed for mounting comes with extendable legs to support the weight. Or the monitor is designed on some kind of lever system that can be set to balance. Lots of options.
tpmx 372 days ago [-]
I kind of like your first idea. The motherboard would be vertically mounted, and then have 90 degree PCIe risers that are structurally reinforced somehow, so that the heavy GPU can also be mounted vertically. You could have 2 of them, one for the GPU and one for additional cards. The whole thing would be mounted via the VESA screw holes on the monitor.

Would probably mostly work for people with quite large monitors though. 40+ inch 4k monitors rock!

... or we just stick with tower cases and some kind of structural reinforcements for the heavy GPU.

dragontamer 373 days ago [-]
1. While I agree we're beginning to reach absurd proportions, lets really analyze the situation and think about it.

2. Are there any GPUs that actually have performed physical damage on a motherboard slot?

3. GPUs are already 2-wide by default, and some are 3-wide. 4-wide GPUs will have more support from the chassis. This seems like the simpler solution, especially since most people rarely have a 2nd add in card at all in their computers these days.

4. Perhaps the real issue is that PCIe extenders need to become a thing again, and GPUs can be placed in an anchored point elsewhere on the chassis. However, extending up to 4-wide GPUs seems more likely (because PCIe needs to get faster-and-faster. GPU-to-CPU communications is growing more and more important, so the PCIe 5 and PCIe 6 lanes are going to be harder and harder to extend out).

For now, its probably just an absurd look, but I'm not 100% convinced we have a real problem yet. For years, GPUs have drawn more power than the CPU/Motherboard combined, because GPUs perform most of the work in video games (ie: Matrix multiplication to move the list of vertices to the right location, and pixel-shaders to calculate the angle of light/shadows).

eigenspace 373 days ago [-]
> 2. Are there any GPUs that actually have performed physical damage on a motherboard slot?

I have a MSI RTX 3070 2 fan model. It hasn't damaged my PCI-E slot (I think), but it's weight and sag causes some bending that now makes it so that my fan bearing makes a loud noise if spun up high.

My solution has been to turn my PC case so the motherboard is parallel to the ground and the GPU sticks straight up, eliminating the sag. Whisper quiet now.

If this is happening with my GPU, I shudder to imagine what it's like with other GPUs out there which are much bigger and heavier.

pdimitar 373 days ago [-]
Yeah, I do that for a long time. Had a few accidents where I panicked until it turned out the card was just sagging and the PCIe connection was being stopped.

After it happened the 3rd time I just cleaned up a little space and put the PC lying on its side. Zero problems since then.

test6554 373 days ago [-]
moonka 373 days ago [-]
I don't even understand what's supposed to be funny about this joke.
sk0g 373 days ago [-]
I suppose you are not mysoginistic enough for the joke. I'd say that is a positive.
pcdoodle 372 days ago [-]
I laughed.
moonka 372 days ago [-]
Can you explain it? I think I'm missing something.
BoorishBears 373 days ago [-]
I've seen a lot of people ignore which screws they're using to retain their GPU

The screw should have plenty of surface area to properly secure the card. You'll still have _some_ sag, but my 3 pin 3090 doesn't sag half as badly as examples I've seen of much smaller cards

fjert 373 days ago [-]
I have an EVGA 3070 and also had the sag issue. My case came with a part to support the GPU though, but I didn't realize until I solved it another way: I just doubled up those plates that you screw the GPU into so there was no way it could pivot and sag.
topspin 373 days ago [-]
> 2. Are there any GPUs that actually have performed physical damage on a motherboard slot?

Yes. I've seen both a heavy GPU and an HSM card damage a slot. One happened when a machine was shipped by a commercial shipper. The other happened when the machine was moved between residences. It doesn't occur to people that the mass of a card swinging around is a problem when the case is moved.

The HSM one was remarkable in that it was a full length card with a proper case support on both ends.

Also, this isn't just about damaging the PCI-E slot. Heavy cards bend PCBs (both the MB and the dangling card) and bending PCBs is a bad idea: surface mount devices can crack, especially MLCCs, and solder joints aren't flexible either. No telling how many unanalyzed failures happen due to this.

If you have a big GPU don't let it dangle. Support it somehow.

Another area where the conventional layout is struggling is NVME. They keep getting hotter and so the heatsinks keep getting larger. Motherboard designers are squeezing in NVME slots wherever they can, often where there is little to no airflow...

Arrath 372 days ago [-]
> One happened when a machine was shipped by a commercial shipper. The other happened when the machine was moved between residences. It doesn't occur to people that the mass of a card swinging around is a problem when the case is moved.

Huh. Good point. I'll be moving soon, and have kept the box my case came in as well as the foam inserts for either end of the case. I might just remove the GPU and put it back in its own box for the move, as well. Thanks for bringing that up.

pohuing 373 days ago [-]
> If you have a big GPU don't let it dangle. Support it somehow.

Or remove it. Takes five minutes to unplug and unscrew a card. Had to do that with my 6900xt recently while moving.

Macha 373 days ago [-]
2. We crossed that point generations ago. High end GPU owners are advised to remove GPUs from their system before transporting it and PC communities get posts often enough of people who had consequences from not doing so. Over a longer term even a lighter card can deform motherboards - I had a 290x which did so to an old Z87 motherboard over 5 years with the result the motherboard was no longer flat enough to mount the backplate of a replacement cpu cooler to.

4. Don't forget high end GPUs are also getting longer, not just thicker. So increasing sizes both give and take away weight advantages

kenhwang 373 days ago [-]
> PCIe extenders need to become a thing again

PCIe extenders are a thing already. Current PC case fashion trends have already influenced the inclusion of extenders and a place to vertically mount the GPU to the case off the motherboard.

GPU sag is also a bit of a silly problem for the motherboard to handle when $10 universal GPU support brackets already exist.

jrockway 373 days ago [-]
I have one of these for a much smaller card, mostly so that cold airflow from the floor intake fans actually has a path to get up to the RAM and VRMs. This is a workaround for a case that doesn't have front intake, which is preferable in my opinion.

It does look a little cool, but I always worry a little about the reliability of the cable itself. Does it REALLY meet the signal integrity specifications for PCI-E? Probably not. But, no unexplained crashes or glitches so far and this build is over 2 years old.

luminiferous 373 days ago [-]
LTT has a video where they tried to see how many PCIe riser cables they could string together before it stopped working.[1] They got to several meters. Maybe you could argue that it's worse inside a PC case since there's more EMI, but it seems like your PCIe riser cable would have to be very out of spec before you'd notice anything.

[1] https://www.youtube.com/watch?v=q5xvwPa3r7M

terom 373 days ago [-]
I wonder if that benchmark actually loaded the PCIe bus to any significant degree after the initial benchmark startup, or just updated a couple small parameters on a single scene and thus mainly just tested the local computation on the GPU?

You'd want to somehow monitor the PCIe bus error rate - with a marginal signal and lots of errors -> retries, something that loads the bus harder (loading new textures etc) could suffer way more.

They do briefly show a different PCIe riser made out of generic ribbon cable [1, 3:27], and say that one failed after chaining only two ~200mm lengths. The quality of the riser cable certainly matters.

[1] https://youtu.be/q5xvwPa3r7M?t=207

Godel_unicode 373 days ago [-]
You need Steve for that kind of testing, LTT would be busy putting rgb on it and then (badly) water cooling it so they could sell you a backpack with no warranty.
userbinator 373 days ago [-]
It's not clear whether they reached a limit of drive strength or latency (I doubt EMI is the factor, since he said those are shielded) but that's a good demonstration of the resiliency of self-clocked differential serial (and aggregated serial) buses. The technology is much closer to networking standards like Ethernet than traditional parallel PCI, with features like built-in checksums, error correction (newer versions), and automatic retransmit.
willis936 373 days ago [-]

The 650 Ti is PCIe 3.0. PCIe 4.0 doubles the bandwidth. PCIe 5.0 doubles the bandwidth again. The RTX 40 series GPUs still use PCIe 4.0, which have commonly available conformant riser cables. I suspect the story for PCIe 5.0 will be different.

jrockway 373 days ago [-]
Oh yeah, I've actually seen that video. I guess I have no reason to be even slightly paranoid!
MBCook 373 days ago [-]
My proposal isn’t too different. Move the one ultra-fast PCIe slot to the top of the motherboard. It would be mounted so the GPU plugs in parallel to the motherboard, on the same plane, above the motherboard on a tower case. The few other PCIe slots that exist can stay at the bottom.

Only downside is the case has to be taller. Not sure if that would be considered a problem or not.

This doesn’t really help dual GPU setups, but those have never been common. I don’t have a good solution there. I guess you’re back to some variation of the riser idea.

c_o_n_v_e_x 373 days ago [-]
Not a bad idea however you have hard limits on how physically long the PCIe lanes can be. We had problems making sure we hit signal budget for a PCIe gen 4 slot on an ATX motherboard. The problem (PCIe lane length) gets worse as the speeds increase.
MBCook 373 days ago [-]
My hope was it could similar length to today, just on the other side of the CPU.

Thinking about it though isn’t that where a lot of motherboards put a lot of their power regulation circuitry? I’m sure something would have to move.

But excellent point. That could easily bet a much bigger sticking point that I didn’t really consider too much.

zapdrive 373 days ago [-]
On a further note, why does it even have to be inside the case? Make a slit in the case on the top, so that the PCIe slot is sticking out. Stick a GPU in that slot, supported by the case. The GPUs these days look much cooler anyways.
Godel_unicode 373 days ago [-]
Briefly, EMI.
topspin 373 days ago [-]
> Only downside is the case has to be taller

Not necessarily. A flexible interconnect would allow the GPU to be planar with the MB; just bend it 180 degrees. Now your GPU and CPU can have coolers with good airflow instead of the farcical CPU (125mm+ tall heatsinks...) and GPU cooling designs (three fans blowing into the PCB and exhausting through little holes...) prevailing today.

My idea is to separate the whole CPU complex (the CPU and heatsink, RAM slots, VR, etc.) from everything else and use short, flexible interconnects to attach GPUs, NVMe and backplane slots to the CPU PCB, arranged in whatever scheme you wish.

MBCook 373 days ago [-]
I was kind of hoping doing it that way would let you put big CPU style coolers on the GPU parts with a lot more height than a 1x or 2x expansion slot.

If you “folded” the GPU over the CPU to save height I would think that would be worse than today for heat.

Maybe I’ve got this backwards. Give up on PCIe, or put it above the rest of the motherboard. The one GPU slot, planar to the motherboard, stays below. Basically my previous idea flipped vertically.

The other PCIe slots don’t need to run as fast and may be able to take the extra signal distance needed. The GPU could secure to the backplane (like my original idea) but would have tons of room for cooling like the CPU.

topspin 373 days ago [-]
> If you “folded” the GPU over the CPU to save height I would think that would be worse than today for heat.

Why? Cooling would be far better: the CPU and GPU heatsinks would both face outward from the center and receive clean front-to-rear airflow. Thus, looking down from above:

        intake air -->  [cpu heatsink]    exhaust --> 
                    | <-- CPU/GPU interconnect
        intake air --> [gpu  heatsink]    exhaust -->

The power supply and power connectors are on the bottom. Another PCB lays flat on top to host open PCI-E slots, NVMe, whatever, connected 90 deg. to the CPU PCB with one PCI-E slot interconnect. All interconnects are short. Air flow is simple and linear. The CPU/GPU Heatsinks are large and passive: you only need intake fans.
rawoke083600 373 days ago [-]
Upvote for ascii-visual-aids
topspin 372 days ago [-]
Thanks. :)

I've been refining this. I'm actually learning FreeCAD to knock out a realistic 3D model.

One obvious change: run the CPU/GPU interconnect across the bottom: existing GPU designs could be used unmodified (or enhanced with only a new heatsink) and the 16x PCI-E lanes for the GPU would be easier to route off the CPU PCB.

There are virtually no significant differences between the motherboard layout IBM promulgated with original IBM PC (model 5150) in 1981 and what we have today. That machine had a 63W power supply and no heatsinks or fans outside the power supply. The solution to all existing problems with full featured, high power desktop machines is replacing the obsolescent motherboard design with something that accommodates what people have been building for at least 20 years now (approximately since the Prescott era and the rise of high power GPUs.)

whywhywhywhy 372 days ago [-]
>This doesn’t really help dual GPU setups

They are with the people who spend the most money on GPUs

Shorel 372 days ago [-]
Are they?

Seems with the last GPU generation most enthusiast websites recommend to go for a bigger card instead of going SLI.

whywhywhywhy 372 days ago [-]
GPGPU workflows scale in a linear fashion so the more you buy the faster your workloads run
RL_Quine 373 days ago [-]
Yeah, I have a motherboard with a bent out of shape port because of the weight of the card in it. My current desktop has a sag of easily half an inch at the end of the card and it’s not even a particularly large one by current standards. The ATX spec obviously wasn’t designed with GPUs this heavy and this power hungry.
somat 373 days ago [-]
historically cases had a bracket at the front to support full length cards. I even remember I once had a reference amd card that had an odd extension so that it would be supported by the forward full length brackets.

I have to admit I have not seen that front bracket for a long time. some server chassis have a bar across the top to support large cards. this would bet great except gfx card manufacturers like to exceed the pci spec for height. that bar had to be removed on my last two builds. now days I just put my case horizontal and pray.

techdragon 373 days ago [-]
I came here to mention the front support bracket. You'll find it on the larger AMD workstation cards more often than others, I first remember it on a FirePro reference card, and some searching turned up examples of it for the AMD FirePro V7900, and a few other models.

I've also had the vertical clearance issue since I try to rack mount all my gear now I've got a soundproof rack, its very annoying to need more than 4U of space just to fit a video card in a stable vertical orientation.

7speter 373 days ago [-]
Gpu sag is a big issue in gaming computers. I had a small (by comparison and of more contemporary graphics cards) rx480 and bought a cheap 10 dollar graphics card brace to reduce its strain on the pci slot in 2021 to help reduce its chances of failure during the shortage. I use the brace to hold up my new ampere card now (which is maybe twice the length of the rx480).
CydeWeys 373 days ago [-]
> Are there any GPUs that actually have performed physical damage on a motherboard slot?

It's quite common to suffer damage from movement, especially in shipping, to the point where integrated PC manufacturers have to go to great lengths to protect the GPU in transit.

throwmybro123 373 days ago [-]
I’m not sure how the author, who programs GPUs, doesn’t comprehend the cooling is part of market segmentation. A 3090 Turbo is a two slot solution, but NVIDIA forced vendors to discontinue it to prevent creep into datacenters.

And I’m sure the licensing bros will come out and shout about licensing or something irrelevant. My dudes, I operate 3090s in the data center, it saves boatloads of money for both upfront, licensing, power and therefore TCO, and fuck NVIDIA.

superchroma 373 days ago [-]
RE 2: We were shipping a system with a 980ti and it ripped the PCIE slot off the board despite being propped with foam.
teaearlgraycold 373 days ago [-]
Hard foam? Regardless I'd recommend against shipping with a CPU cooler or GPU installed but some SIs seem to get away with it with self-fitting hard foam packs.
ikiris 373 days ago [-]
... maybe don't ship a built system then? Those boxes get dropped
CydeWeys 373 days ago [-]
That's not a reasonable answer. Many people want to buy assembled systems, and lack the skills/inclination to do it themselves. Pre-built systems is a huge $1B+ market, and "kill the market entirely" is not an acceptable answer.
ikiris 373 days ago [-]
Your business model is not required to be viable.

I said don't ship them. I did not say don't build prebuilt systems.

CydeWeys 369 days ago [-]
This reply doesn't make any sense. It is a viable business model. It has total revenues across all the major players of billions of dollars per year.

And how the hell do you not ship them? You're not making any sense here. There's no alternative to not shipping them, not unless you're planning on having system builders show up individually to clients' houses and assemble PCs on the spot. That business model is way less economically efficient than simply assembling PCs centrally and accepting some breakage in shipment.

bearmode 373 days ago [-]
Breakage is normal in any industry, what are you suggesting as an alternative? There are companies shipping nationally and internationally with very few issues
sidewndr46 373 days ago [-]
Dropped? The UPS guy regularly throws stuff from the curb and I hear it bounce off my front door then land on the front porch.
superchroma 373 days ago [-]
We didn't have a choice. We were building to a strong spec and sending a unit to trade shows.
Dylan16807 373 days ago [-]
That's what internal screws and external padding are supposed to be able to compensate for.
Tsiklon 373 days ago [-]
Workstation cards are traditionally supported on 3 sides so don’t suffer this sag. In such systems there’s usually an extension bracket screwed into the card to increase the length, allowing it to reach a support slot to hold it steady.
cyanydeez 373 days ago [-]
In reality, there is no reason GPUs should be add-ons.

Just star designing motherboards with integrated GPUs of sufficient size.

Case closed.

CydeWeys 373 days ago [-]
The GPU is the most expensive component in gaming PCs these days, thus it makes the least sense for it to be the hard-wired component as there is the most price diversity in options. I have definitely upgraded GPUs on several computers over the past decade and I'm very thankful I didn't have to rip out the entire motherboard (and detach/reattach everything) to do so.

It's only the cheap components without a wide variety of options that make sense to build in, like WiFi, USB, and Ethernet.

NavinF 373 days ago [-]
Note this whole discussion is in the context of the 4090. If you're an enthusiast, soldering the GPU to the mobo forces you to spend $200-$700 more every time you upgrade your GPU because you also have to buy a new mobo and possibly a new CPU if the socket changed.

The GPU is also one of the easiest components to swap today. That's not something I want to give up unless I see performance improvements. Cooling doesn't count because I already have an external radiator attached to my open-frame "case".

I went through 3 GPUs before changing motherboards and I'm still bottlenecked on my 3090, not my 5800X3D. After I get a 4090, I expect to upgrade it at least once before getting a new mobo.

Arrath 372 days ago [-]
Having had a few GPUs go bad on me over the years, I would hate to have to disassemble the entire thing to extricate the mobo/gpu combo for RMA'ing, rather than just removing the GPU and sending it off.
wruza 373 days ago [-]
The main reason is that CPUs get old much slower than GPUs, but “current” socket generations change quickly. Another reason is a combinatorial explosion of mobo x graphics card features, which is already hard.
7speter 373 days ago [-]
Gpus get old slowly too, but youtube influencers convince gamers their current card isn’t enough often ans effectively (I’m not excluded from this group)
stjohnswarts 373 days ago [-]
not that simple. lots of people upgrade their card once or twice before they get a new motherboard/cpu combo.
matja 373 days ago [-]
Wouldn't that be a cause of even more e-waste?
373 days ago [-]
373 days ago [-]
bearmode 373 days ago [-]
Then you lose modularity, which is a huge benefit to PC gaming? Now if you want to upgrade to the newest graphics card, you also need to get a new motherboard. Which also could mean you need a new CPU, which also could mean you need new RAM.

Right now you can just switch the graphics card out for another and keep everything else the same.

danaris 373 days ago [-]
This is already happening (as was noted in other comments on this article).

One of the most prominent examples is the entire Apple Silicon lineup, which has the GPU integrated as part of the SoC, and is powerful enough to drive relatively recent games. (No idea just what its limits are, but I'm quite happy with the games my M1 Max can play.)

sfmike 373 days ago [-]
With mini cube PCs growing in popularity the future will probably be this, a mini PC with every part a USB type modularity for any ram or GPU or hd into that stock cube PC.
YurgenJurgensen 373 days ago [-]
Please no. USB latencies are measured in microseconds. DDR3 latencies are measured in nanoseconds. "USB Ram" would be a disaster.
zaptheimpaler 373 days ago [-]
How about we move to external only GPUs with huge connectors? If GPUs are half the size, power consumption and price of a PC now, they might as well be a separate device. As a bonus the rest of the motherboard & PCs actually get much smaller. A PC without any spinning disks could conceivably just be the size of a NUC by default, something you can travel with when you don't need the beefy GPU.
spartanatreyu 373 days ago [-]
The most expensive part of GPU operation (power-wise) isn't computation, it's actually moving information around.

Moving the GPU further away from the CPU and storage is going to lead to worse latency and power requirements.

Over the next 10-15 years, we'll probably see CPU+GPU packages become mainstream instead of CPU (with basic integrated graphics) and a separate GPU.

userbinator 373 days ago [-]
The most expensive part of GPU operation (power-wise) isn't computation, it's actually moving information around

I doubt that. Compare the GPU temperature (a good proxy for power consumption) when playing a game or doing GPGPU stuff vs playing a video (without GPU acceleration, so it's just acting as a framebuffer). The former involves far more computation, and gets the GPU much hotter.

spartanatreyu 373 days ago [-]
No. There was an internal nvidia presentation from a few years ago that stated that moving data was the hard part. (I can't find the presentation any more, but if anyone can find it, please post it below.)

Previously graphics cards were essentially designed with a single area of the card handling computation, and another area holding memory. Data would need to be moved from the memory, to the computation area, then back again if there were changes that needed to be stored.

As the computation and the memory demands became larger, those areas had to handle more, but so did the bus between those two areas. What was a negligible overhead for the bus became more pronounced as it had to handle more data.

Eventually the energy overhead of transporting that much data across that distance started constraining what was possible with graphics cards.

That is why graphics card architectures have shifted over time to place memory cache units next to computation units. The less distance the data needs to travel, the smaller the power requirements. It's also led to the investment and adoption of stacked memory dies (why grab data from 5cm away in the x-y plane when we can stack memory and grab data 5mm away in the z-direction).

TomVDB 373 days ago [-]
Moving around data is indeed a major issue for any throughout oriented device. But for a gaming GPU, PCIe BW has never been an issue in any of the benchmarks that I’ve seen. (Those benchmarks artificially reduce the number of PCIe lanes.)

In fact, the 4000 series still has PCIe 4.

Moving data around for a GPU is about feeding the shader cores by the memory system. PCIe is way too slow to make that happen. That’s why a GPU has gigabytes of local RAM.

rodgerd 373 days ago [-]
Given that every Mac, AMD-based XBox and Playstation, and a chunk of the Ryzen market are shipping decent unified CPU/GPU combos, I would say that day is mostly here.
ekianjo 373 days ago [-]
> Over the next 10-15 years, we'll probably see CPU+GPU packages become mainstream instead of CPU (with basic integrated graphics) and a separate GPU.

its already the case with arm socs.

bradleybuda 373 days ago [-]
But home PC users / high end GPU buyers (rightly or wrongly) are fanatical about modularity. The server market will likely (has already?) shifted.
calessian 373 days ago [-]
I’m inclined to say that’s more due to integrated gpus not being a good replacement if you need CUDA or play games; If a better combined alternative existed (and not at an outrageous price) I don’t see the majority skipping it die to lack of being modular.
bhedgeoser 373 days ago [-]
I think what the OP is suggesting is that PCs should have external PCIE connectors for plugging in GPU modules.
bfgoodrich 372 days ago [-]
chucklenorris 373 days ago [-]
The problem with external gpus is that the pcie5x16 slot requires signal integrity chips every 5 inches or so [1]. Even for pcie4 it's bad, many people had blue screens or boot issues when using pcie3 only rated riser cables with a pcie4 video card, even though electrically they have the same connections. So, having a huge cable with x16 lanes of pcie5 on the back of the computer doesn't seem that feasible. Maybe fiber optics could be a thing.

[1] https://www.planetanalog.com/signal-integrity-challenges-set...

Bluecobra 373 days ago [-]
Fiber optic would have higher latency vs. copper. Not sure how much of a difference that would make on a bus but I would assume the timing is pretty tight.
bhedgeoser 373 days ago [-]
I remember linuxtechtips using a meter long wire to connect a 3090 to a computer and it worked flawlessly.
ChoGGi 372 days ago [-]
3 metres, not to mention it wasn't a single cable, but a bunch hooked up to each other. Though I believe that was PCIe 3, so 4 might be more picky.
jvanderbot 373 days ago [-]
You can easily buy a NUC ECE (size of a graphics card) and use a pcie cable to connect to an actual card. This is my plan for my next machine.
pdimitar 373 days ago [-]
How would that work? I admit I never even thought of PCIe cables, thought it was always just a physical slot you have to put stuff in.
nighthawk454 373 days ago [-]
there are PCIe riser cables that you can imagine as an extension cord for a PCIe slot. they're popular in some thin server chassis or in miniITX builds where the form factor dictates the GPU must be in a certain orientation to fit.

there's also the method of external GPUs via Thunderbolt (which carries PCIe) that is more practical for putting a GPU in an entirely separate enclosure

imtringued 372 days ago [-]
Ok so just plug your computer into the GPU ;)
pubby 373 days ago [-]
I wonder how far that connection could be before it stops working. Probably not much!
jacooper 373 days ago [-]
Here is an LTT video about how many PCI-E extensions one can daisychain without problems.

The limit is around 3 meters.


bmicraft 373 days ago [-]
One should note that it will works way worse with higher pcie versions of today
Red_Leaves_Flyy 373 days ago [-]
Are you referring to pcie 4?
layer8 373 days ago [-]
The RTX 4000 series uses PCIe 5, and the article references PCIe 6.
TomVDB 373 days ago [-]
I was wondering about that. According to Nvidia RTX 4090 product page, it still used PCIe 4.
jacooper 372 days ago [-]
Yup, it only uses PCI-E5 power connectors. The GPU is gen4.
bhedgeoser 373 days ago [-]
A meter of extension adds ~10ns of latency (speed of light).
arvinsim 372 days ago [-]
Isn't this what the thunderbolt enclosures with GPUs are trying to do?
bofh23 371 days ago [-]
I’m pretty sure Thunderbolt only has the equivalent of 4 lanes of PCIe 3.0 while GPU cards tend to be 16 lane.
373 days ago [-]
winkeltripel 373 days ago [-]
Could we just follow the atx spec? There is a maximum length for expansion cards, and at that end there are optional supports. These are in servers already. Just start using all that case volume to support the GPU.
somat 373 days ago [-]
Cheers, bravo, exactly what I wanted to say. However just to be pedantic, I just read the atx spec to find this exact information and it is not there. I suspect the expansion card dimensions are in the pci spec.
Nition 373 days ago [-]
Is that the same as the one shown in the article's spec diagram, or something different again?
Lammy 373 days ago [-]
My case has these and I would love to know the proper keywords to find extension brackets I could add to my shorter cards. Searching for "extender" returns electrical extenders for the PCIe port itself.
MBCook 373 days ago [-]
I was always under the impression that was to help stabilize long cards so they don’t sag or fall during transport.

But these aren’t just long cards. They have huge heavy chunks of metal and fans on them in the name of cooling.

Are those brackets strong enough for the job? I remember them being basically shallow plastic slots.

stjohnswarts 373 days ago [-]
there are $10 brackets that solve all this for most people, not sure why there is an HN panic to suddenly rearchitect the whole PC gaming platform.
MrFoof 373 days ago [-]
I made this GIF to illustrate the point of how large these new high-end NVIDIA Lovelace consumer GPUs are: https://i.imgur.com/327INxU.gif

This is the ASUS RTX 4090 ROG STRIX. Air cooled, no waterblock. That is a mini-ITX form factor motherboard, hence why it looks so comically large by comparison.

This is one of the physically smallest 4090s launching. Its confirmed weight is 2325g, or 5 ⅛ lbs. Just the card, not the card in its packaging.

chx 373 days ago [-]
There used to be so called PIO motherboards from China. These were slightly larger than ITX and the PCIe connector was 90 degrees rotated so the video card was planar with the motherboard. https://imgur.com/a/ve1T0dE

And if we are to reform our computer chassis anyways, we could move the PSU to straddle the motherboard and the video card and even have the VRM inside. High amperage "comb" connectors exist and VRM daughtercard motherboards existed https://c1.neweggimages.com/NeweggImage/productimage/13-131-... Change the form factor so two 120mm fans fit, one in front, one in the back.

So you would have three 120mm front-to-back tunnels: one for the video card, one for the PSU, one for the CPU.

MuffinFlavored 373 days ago [-]
lightedman 373 days ago [-]
Voltage Regulator Module.
SanjayMehta 373 days ago [-]
New design: Switch things around and stick the CPU into a slot on the GPU.
LeoPanthera 373 days ago [-]
Weirdly, this is the design of the Raspberry Pi, wherein the GPU bootstraps the system and brings the CPU up. Though it is a somewhat academic distinction, since both live inside the same chip.
carabiner 373 days ago [-]
The GPU should be the motherboard, and everything sticks onto it.
ungamedplayer 373 days ago [-]
This is gpu manufacturing "wet dream" endgame
DaiPlusPlus 373 days ago [-]
So how would multi-GPU setups work?
rocket_surgeron 373 days ago [-]
Multi-gpu setups for gaming are dead.

Multi-gpu setups for computation could have two SKUs one "motherboard" SKU and one connectorized SKU with the connector NOT being PCIe (after a transition).

They already do multi-form-factor, PCIe and OAM for AMD, PCIe and SXM for Nvidia.

Just drop PCIe, have a large motherboard SKU with a CPU slot and some OAM/SXM connectors in quantities up to what a wall socket can supply in terms of wattage (so, like 1 or 2 lol).

Vestigial PCIe cards can hang off the CPU card, if they're even needed.

High speed networking and storage is already moving to the DPU so these big GPUs, unhindered by the PCIe form-factor, could just integrate some DPU cores like they do RT cores, and integrate high speed networking and storage controllers into the GPU.

Home user? You get 1 or 2 DPU cores for NVMe and 10-gig Ethernet. Datacenter? 64 DPU cores for 100-gig and storage acceleration. Easy-peasy.

kuschku 373 days ago [-]
> what a wall socket can supply in terms of wattage (so, like 1 or 2 lol).

So, you expect GPUs to take 1500W of power, each? (230V @ 16A)

rocket_surgeron 372 days ago [-]

The motherboard and then 1 or 2 expansion sockets, for a total of 3.

3 x 450W for GPU (Nvidia says 4090 draws 450W-- I think they're lying and will wait for reviews to see the truth) and 500W for rest of system. Though that might be a bit low, the 11900K has been measured at what, 300W under load? And you'd need a budget for USB-C power, multiple NVMe, fans, and whatnot. Maybe the spec would accommodate high-core-count enterprise processors so 600W+ would be wiser.

Even in euroland 2KW, which is what a maxed out system would be at the wall socket, is a bit much. They don't even allow vacuum cleaners over 900W to be sold.

kuschku 371 days ago [-]
2kW isn't actually that much — you can buy ATX power supplies up to 3.5kW for usage in 230V countries. Often used for mining, but also useful if you want to accelerate your blender renders by just throwing multiple GPUs at the task.
imtringued 372 days ago [-]
Honestly just use chiplets technology for more cores and get rid of the multi GPU concept entirely.
DaiPlusPlus 372 days ago [-]
Multi-GPU isn’t just about single-application performance.

Multi-GPU is necessary if you need more display output ports: with limited exceptions every GPU I’ve seen is 3xDP+1xHDMI or worse. While a single DP can drive multiple monitors it limits your max-res/refresh-rate/color-depth - so in-practice if you want to drive multiple 5K/6K/8K monitors or game at 120Hz+ you’ll need two or more cards, with-or-without SLI/etc.

midoridensha 373 days ago [-]
You buy a motherboard with the number of GPUs you want on it.

Or maybe they could devise some sort of stacking system, with each GPU board separate and stacked.

DaiPlusPlus 373 days ago [-]
> Or maybe they could devise some sort of stacking system, with each GPU board separate and stacked.

Yeah, I remember those stackable/modular computer concepts that industrial-design students loved to put in their portfolios from the late-1980s to the mid-1990s; I get the concept: component/modular PCs are kinda like the modular hi-fi systems of the 1980s, except those design students consistently failed to consider how coordinated aesthetics are the first thing to go out the window when the biz-dev people need the hardware folks to save costs or integrate with a third-party, etc.

...it feels like a rare miracle that at least we have 19-inch racks to fall-back on, but those are hardly beautiful (except the datacenter cable-porn, of course):



373 days ago [-]
c_o_n_v_e_x 373 days ago [-]
You just re-invented COM-HPC / COM-Express... kinda.
justsomehnguy 373 days ago [-]
Back to the roots, baby: https://www.biocomp.net/o91474.htm
probably_a_gpt 373 days ago [-]
Give it a few years
hakfoo 373 days ago [-]
This is (much less of a) problem on a flat layout, like what used to be called a "desktop" case, instead of the conventional tower. Then the weight of the card is just pushing straight down in the direction the card already wants to be oriented.

I'm using a pretty heavy modern GPU (ASRock OC Formula 6900XT) in a Cooler Master HAF XB with that layout, and sagging and torquing is not much of a concern. The worst part is just fitting it in, since there's like 2mm between the front plate and the board-- you have to remove the fans so you can angle the card enough to fit.

I also suspect that if we went to the 80's style "a full length card is XXX millimetres long, and we'll provide little rails at the far end of the case to capture the far end of a card that length" design, it would help too, but that would be hard to ensure with today's exotic heatsink designs and power plug clearances.

alkonaut 372 days ago [-]
Why not just make an ATX GPU with a CPU slot on it? With the size and cost of these things, it's the rest of the machine that feels like a peripheral and not the VGA.

The GPU is the main part of the machine by cost, weight, complexity, power consumption. And it's not even close.

db48x 373 days ago [-]
At this point both the motherboard and graphics card need to be mounted on the back plate of the chassis, so that they can both use tower coolers. You can already use a PCIe extender to achieve this, but it should become the standard.
wtcactus 373 days ago [-]
Although the idea of the author focus solely on how to fit/support the card in the motherboard, and to provide an existing cooling solution, I actually find it a bit too much, this race to higher performance based on increasingly higher power requirements.

New NVIDIA cards will draw 450W, and, even if you lower that in settings, the all package will still need to be manufactured to support those 450W at various levels.

I wonder what are games doing that require that extra power, seriously. I, personally, would much prefer to slightly have to lower settings (or expect devs to take at least some basic steps to optimize their games) than have a 450W behemoth living inside my computer.

Meaning, 40xx series will be an obvious pass for me. My 1080 Ti is actually still great in almost all aspects.

intrasight 373 days ago [-]
It's the IBM PC legacy. The won and we've lived with that form-factor now for 40 years. A new PC looks very much like that of 1982. Back in '82 when I started in robotics tech, we mostly used VME. A super-robust interconnect platform. There is no "motherboard" with VME and similar bus architectures. There is a backplane. Why can't we have the benefits of a physical platform like VME but with the electrical form-factor of PCIe?
Tuna-Fish 373 days ago [-]
An important reason why you cannot have that is that the VME is a shared bus, while PCIe is a very high-frequency LVDS point-to-point system that you absolutely cannot implement on a shared bus.
rocket_surgeron 373 days ago [-]
VPX (VITA 46).

Think of it like VMEexpress.

I use PCIe over VPX every day.


Please, lord, somebody else outside of aerospace use it so more vendors get involved and availability and basically everything about the entire supply chain gets better through competition.

imtringued 372 days ago [-]
I always thought VPX is cool but then again there are no consumer products that use it so it is basically irrelevant.
willis936 373 days ago [-]
There are fabric backplane solutions such as fibre channel and microtca (leverages pcie or ethernet).

There's no reason you can't get benefits from a fabric over point-to-point today, it's just big and expensive.

intrasight 373 days ago [-]
Like I said - not the electric part but the mechanical part.
dragontamer 373 days ago [-]
I had a professor talk about those old systems once.

Weren't those systems a bus architecture? Like the backplane wires all shared the same wires so it was possible to keep stringing them along?

Modern PCIe, and even USB3, are star networks. Everything directly connects to the northbridge like an Ethernet cable star network connects to a switch.


PCIe slots communicate the idea well. There are only so many dedicated wires the northbridge exposes to USB3 or PCIe.

This, it makes sense to have those limited wires in the form of PCIe slots or USB3 connectors.

wmf 373 days ago [-]
A backplane and a bunch of additional connectors add cost that the PC market doesn't want to pay.
techdragon 373 days ago [-]
checks features of motherboard ... additional connectors, yep... PCIe backplane, yep...

Having bought both... it is definitely costs they don't want to pay, but not the cost of those features. It's more like industrial grade PSUs and redundancy that get mandated by various specifications and end up reducing the market size which further racks up costs.

wmf 373 days ago [-]
PCs could have their own backplane standard that's higher-volume and doesn't require industrial-grade anything, but it would still be too expensive. They sweat every yuan.
techdragon 373 days ago [-]
My point was that the economics of higher-volume manufacturing result in eventually subtracting everything to the point that the "backplane" is the motherboard.

Early ISA backplanes were passive and you basically just had a PC on an ISA card plugged into the ISA bus, with maybe some extra connectors for power and peripherals. Then PCI backplanes started out this way but quickly gathered enhancements that resulted in the "System Host Board" interface and a special connector for more power and IO to/from the "CPU card" in order to do things like support ISA, PCI, peripherals, and later PCI-X slots. Rather than having the extra complexity of the connectors, the high volume PC market folded all of this "onto the backplane" creating the modern motherboard. It might not be clear unless you know a bit about early computer backplanes like S-50/S-100, STD bus, STEbus, VME bus and the equivalent Passive ISA backplanes... by the 90s this is well underway and we already have motherboards that you could recognise as "a motherboard" and "backplanes" that are noticeably different, having evolutionarily diverged during the 80s under differing economic pressures.

virgulino 373 days ago [-]
I liked the new 4090. 12 fans seems reasonable for a sub 1 kW card. Those 2 AC power connectors on the back are a nice innovation. Great benchmark numbers. That they managed to have 21 cards in stock at launch is fantastic!

The 4090 Ti looks fantastic too. Totally worth the risk of fire.


msbarnett 373 days ago [-]
> A 4080 series card will demand a whopping 450 W,

No, that was just a rumour that was floating around. The 4080 16GB model is 340W TGP, the 12 GB is 285W TGP out of the box. The 3080 (10 GB) was 320W TGP, as a comparison point.

Scene_Cast2 373 days ago [-]
Perhaps a typo? The 4090 official TDP is 450W.
msbarnett 373 days ago [-]
Could be. But the pre-launch rumours also claimed the 4080 was a 450 W product, and the 4090 is so niche (basically it’s the old Titan line) that it’s not worth handwringing about because it’s not targeted at normal consumers (for “I’d buy a > $1000 GPU for gaming” values of “normal”).
willis936 373 days ago [-]
Nvidia is attempting to increase sales on the high-end by positioning the 4080 and 4070 (aka 4080 12 GB) at a much higher price point than the market demands while giving good value to the 4090 (60% more performance than the 3090 Ti which launched at $2000).
cabirum 373 days ago [-]
My motherboard says it has "reinforced" pcie slot :)

But yea, I have to say peak power consumption has to be regulated so companies compete in efficiency, not raw power.

gh02t 373 days ago [-]
There sorta is a cap in place given that most residential circuits in the US are 15 or 20 amps. We're already not far off of that being a serious concern even on non-crazy builds. Edit: ok maybe crazy builds but still it's not that far fetched on current trajectories.
MBCook 373 days ago [-]
So 15 A x 120 V is 1800 W. The problem is even in a dedicated circuit you can’t use that continuously, it’s considered dangerous.

For continuous draw you’re limited to 80% of the line’s rating (in the US). People who have looked into EV charging have certainly run into this fact.

So the computer (and sound/monitor/etc) have to stay under ~1400 watts if it’s on a dedicated circuit.

1/3rd of that for just the GPU I’d pretty crazy. I honestly wouldn’t have thought GPUs could get this high.

Also, I remember my computers being hot in a small room long ago. A lot of the space heaters on Amazon are rated either 750 or 1500 W.

I don’t think I wanna be near a computer that hot.

midoridensha 373 days ago [-]
Don't forget the PSU isn't perfectly efficient; it's usually between 80 and 90 percent. So for safety, about 1100W (for the total of the DC power draws of the components fed by the PSU) is really the maximum for continuous draw, and as you pointed out, there's also monitors, sound, etc. A typical monitor now is probably 50W or so (which is actually really good compared to the CRT behemoths of yesteryear).
kenhwang 373 days ago [-]
I don't know if I'd call 1800W (residential circuit limit) builds "non-crazy". Even half that power is rare to see since that's pretty much two top of the line consumer graphics cards paired to a top of the line enterprise CPU.
jquery 373 days ago [-]
Two 3090s plus a 64 core Threadripper pro will get you beyond 1000 watts alone. Don’t forget power for the 512gb of ram (200 watts), the motherboard, the fans and hard drives and accessories and gamer lights (well, gamer lights aren’t necessary). That’s definitely a crazy build though.
jsjohnst 372 days ago [-]
> Two 3090s plus a 64 core Threadripper pro will get you beyond 1000 watts alone. Don’t forget power for the 512gb of ram (200 watts), the motherboard, the fans and hard drives and accessories

Yes, PEAK instantaneous wattage could easily be well over 1kW for a crazy rig like that, but unless you are specifically torture testing the rig over a long duration, it simply isn’t accurate that you are drawing anywhere near the max continuously under gaming conditions.

> Two 3090s plus a 64 core Threadripper pro will get you beyond 1000 watts alone

3090s are 320/350W each and 64 core Threadripper Pro has a TDP of 280W, just as an FYI.

jsjohnst 373 days ago [-]
> We're already not far off of that being a serious concern even on non-crazy builds.

Show me an example of a “non-crazy” build that’s over 1,200W, let alone over 1,500W. I’m genuinely curious.

gh02t 373 days ago [-]
I said near, we still have some breathing room. But as noted elsewhere, 15 A is a max for short periods, the practical limit for continuous use is more like 1200-1400 W which definitely seems like it might be plausible soon.

Plus other things on the circuit. Fancy ultra bright HDR monitors use a decent chunk of power, and anything else.

Sohcahtoa82 372 days ago [-]
1200 W, even for a short period of time, is crazy for a desktop computer.

I have an i9-9900K and an RTX 3080. My computer is also plugged into a Kill-a-watt to measure its power usage.

I ran Prime95's CPU torture test using the max power option while also crypto-mining on the GPU. I peaked at 650 watts for a couple seconds until the CPU began thermal throttling.

If I can't hit 700 watts while trying to use as much energy as possible, I can't imagine what monster system would even touch 1000 watts, assuming we're still talking consumer-grade.

gh02t 372 days ago [-]
Counting my monitors and the small network switch on my desk, I can hit around 750-800 watts sustained with a 5950X and a 3090 running compute intensive workloads as measured at the wall by a power meter.

So yeah, there's plenty of head space but I have also hit circuit breaker issues recently on a proper (non consumer) workstation so the limit is fresh in my mind.

jsjohnst 372 days ago [-]
> But as noted elsewhere, 15 A is a max for short periods, the practical limit for continuous use is more like 1200-1400 W which definitely seems like it might be plausible soon.

Per NEC in the US, the wire used in a circuit has to be rated to handle a current draw of up to 80% continuously at the specified temperature max for the type of installation.

80% of 15 is 12, the average home in the US has split phase incoming power with a nominal voltage of 120V (1440W) with many as high as 125V (1500W). Note that is current draw over long periods of time, which a gaming rig under heavy gaming won’t usually do. Maybe a mining rig back in the day would, but otherwise it’s just not true you’re continuously drawing anywhere near 1,000+ watts on anything but the most crazy setups.

> I said near, we still have some breathing room.

No, you made a wild ass conjecture not based in fact and when gently questioned on it, you shifted goal posts and made up more conjecture.

db48x 373 days ago [-]
I don't really agree that max power should be regulated, but power efficiency should be featured much more prominently in reviews, as should TCO.
7speter 373 days ago [-]
I think this problem will resolve itself. Nvidia pulled out all of the stops with the 4090 because they thought (hoped) they were going to be able to sell it in an insatiable market that was begging for the biggest most powerful die possible, mostly because of mining. Gamers have no appetite for such a monstrosity (the gpu equivalent to national lampoon’s family truckster), and there aren’t really any games on the horizon that requires the 4090’s sheer power. Nvidia is probably going to have no choice but to make smaller, cooler dies with more modest gains relative to lovelace in their next generation because thats what the market will force them to do.
chii 373 days ago [-]
> there aren’t really any games on the horizon that requires the 4090’s sheer power

but with things like stable-diffusion, such a card might have other uses other than games. For example, live paint-ins for photoshop, and other tidbits that are currently not realistically possible for the consumer.

7speter 371 days ago [-]
Sure, but maybe there should be a prosumer line? Or is quadro/titan already that? I just dont know how many gamers are curious about live paint ins for photoshop, or how many gamers can justify *90 costs for video editing and streaming.
Ekaros 373 days ago [-]
Outside certain enthusiast group gamers are somewhat price sensitive. And this new generation went beyond that level. Not yet having anything on acceptable, albeit high price level. But I might be wrong, have to see how sales go after scalpers get burned.
termie 373 days ago [-]
Compare the size of the new air-cooled 40XX cards and the iChill 4090, which is tiny by comparison. The simple answer is just to use liquid cooling if you have a card using 400w. Then all the absurdity just goes away.
RL_Quine 373 days ago [-]
The water cooling on my 3090 is heavier than the air cooling for the same card, as you need a coldplate on both sides of the thing to get rid of the heat.
15155 373 days ago [-]
If you give me a lever long enough and a place to stand, I will move the world.

The length of the 3090 with air cooling is what makes it so heavy on the PCIE connector - localized coldplate weight is not an issue.

RL_Quine 373 days ago [-]
The water block is full coverage, the length of the card on both sides, and is a gigantic block of copper. With the water inside it’s even heavier.
15155 372 days ago [-]
For the 3090 FE, at least, the card is half the length of the air cooler and an EKWB 3090 FE waterblock is the same length as the board.
m463 373 days ago [-]
> Then all the absurdity just goes away.

and once you have liquid cooling, maybe you can go to 800w...

bmitc 373 days ago [-]
What is the end game of consumer GPUs, primarily for gaming? It seems wasteful (?), not sure of the right word here at the moment, to put all this effort into GPUs for general purpose computers and the downstream problems (cooling, space, mounting, etc.) to get arguable improvements in gaming experiences. There seems to be an outright arms race amongst consumers and manufacturers alike for all this, and I personally am not sure why this stuff is so in demand. Are there other consumer markets where high-performance is so accepted and common place?
dotancohen 373 days ago [-]

  > Are there other consumer markets where high-performance is so accepted and common place?
I don't know about today, but a generation of two ago it was street cars, some time before that horses.

Anything measurable will be compared, and people want (more) performance.

bmitc 373 days ago [-]
But performance street cars feel much more niche than gaming rigs. It does feel that performance (or perceived performance) oriented things in the computer market do seem to bleed into the broader market more than other markets. It's an interesting phenomenon.
trashtester 373 days ago [-]
Even "normal" cars tend to come with a lot more hp than most people need most of the time (unless they live in Germany or somewhere else where you can actually go really fast if you want to). Cars with a 100-200hp engine would probably be almost as usable with half the power.
reportgunner 372 days ago [-]
Gaming is becoming more popular and less of a "nerdy" niche thing. Marketing is telling people who are new to gaming that better graphics and more resolution is more gaming so they believe they need to play games in 4k at 240hz.
acapybara 373 days ago [-]
The HGX A100 solves this problem: https://developer.nvidia.com/blog/introducing-hgx-a100-most-...

Basically we need to move away from slots!

lightedman 373 days ago [-]
I'd rather have a dedicated socket on the motherboard where I can just slap in a GPU chip and use whatever RAM I have soldered onto the mobo around the GPU socket.
patates 373 days ago [-]
The GPU should be the motherboard and we should install other lowly components (like the central processor) on top of them.

But seriously, 450 Watts on this day and age of increasing energy prices? Crazy.

teddyh 373 days ago [-]
Stop making cards wider; Bring back full-length cards! With cases having guided slots for them!
db48x 373 days ago [-]
The width of the card is entirely for the heatsink and fanks, not for more electronics.
superchroma 373 days ago [-]
It's not a motherboard problem. How would you integrate support for user provisioned cooling options (to match cooler to the card wattage) and still keep any sort of flexible expansion slot area? GPUs can't be turned into a single chip, there's too much going on, so you're never going to have a CPU cooler situation. So, fine, what if you made them daughterboards mounted like M2 SSD's; that may work, except ATX now has to become ITX to give room for an extra board's worth of space.

It's a PC case orthodoxy issue, really. People want plugs at the back of the box, which dictates how the GPU must sit in the case, and disagreement on GPU sizing means no brackets. Solve these two issues and life gets a lot better.

Or, solve it like SFF case guys solved this problem, by using a PCIE extender cable to allow the GPU to be mounted wherever you like.

mastax 373 days ago [-]
All In One liquid coolers are the answer. They were already getting popular in the RTX 3000 series. They make the card thinner and separate out a lot of the weight. They can't cost much more than some of those massive conventional coolers.
consp 373 days ago [-]
AIO coolers do have the tendency to be hit or miss and can seriously impact performance if done incorrectly. If it's out of the box it should probably be fine though but it's not for everyone to replace a GPU cooler and repaste it.

With these power hungry beasts my custom loop, outside of a tiny case (the radiators make the pc look silly small and are connected to it with quick couplings) suddenly makes practical sense instead of just being a "let's see what I can do" project.

bitxbitxbitcoin 373 days ago [-]
Turn the tower on its side so the motherboard is parallel with the ground and the weight of the GPU keeps it in the PCI-e slot. It is my understanding that GPUs are able to still properly dissipate their heat in this configuration.

Great article!

moandcompany 373 days ago [-]
In the old days, desktop cases existed where you could use the case as a vertical tower or horizontally to put the monitor on top.

People could also move to a rack mountable chassis.

toast0 373 days ago [-]
Watch out or you'll invent desktop cases and we'll put the monitor on top. ;)
lvturner 373 days ago [-]
Came here to somewhat sarcastically write this, glad to see it here.

Less sarcastically, I would quite like to go back to putting my monitor on top of a desktop - sure beats a stack of books.

eigenspace 373 days ago [-]
I do this with my PC unironically.
somat 373 days ago [-]
That is what I started doing. I build my desktop pc's in 4u server chassis[1] I had it vertical in a tower config until a couple months ago when I thought about the dynamics of the situation a bit and moved it to the flat orientation, as you would find it in an actual rack.

1. the cases(even the cheap ones) are better built and have better airflow than most tower cases, as a bonus, no rgb or windows.

puffoflogic 373 days ago [-]
Can you link to the case you use?
somat 373 days ago [-]
This is the one I used on my last two builds, be aware it is a bit of a chonker.

The main reason I choose it over the others is look at all them 5 1/4 bays, the whole front of it is 5 1/4 bays. 5 1/4 bays for days. do you know how many stupid drive bays, fan controllers, sd slots, switches etc you can fit in this case... By my count all of them.



Dylan16807 373 days ago [-]
The one you linked looks like it only has three 5.25 bays at most?
somat 373 days ago [-]
not sure, perhaps(cookies?) you are seeing a different chassis type, the one I ment to link to is "4U, 15 Bays & 8 Fans" it comes with three 3u(5.25) internal drive bays(5 drives each). honestly they suck, pull them to get access to the 5.15 bays.
373 days ago [-]
AmVess 373 days ago [-]
I have a horizontal PC chassis. Easy access to everything and has two large 140MM fans in front. It runs cool and nearly silent. I am never going back to a tower for my main PC again.
0x457 373 days ago [-]
Vertical-mount GPU gang sees no issues.
yarg 373 days ago [-]
Water cooling (with standardised connectors) seems like a sensible option (the heat-sink's the problem, not the GPU).
newZWhoDis 373 days ago [-]
This. Why eject heat into your case?
Havoc 372 days ago [-]
The gamers nexus video is worth a watch for entertainment. The marketing has been cranked up to 11. Graphics cards promising their buyers "absolute dark power". Honestly...
rafale 372 days ago [-]
I actually enjoy the hubris.
prvc 373 days ago [-]
Now that 5.25", 3.5", and 2.5" drives are less common in PCs, it might be time to think of a new standard which puts the GPU in a new form factor which allows for better (or at least simpler) airflow through the case. Seems needless to blow hot air into and out of the GPU just because of where it is situated inside. Imagine only a small number of fans pushing air in a single direction in order to efficiently cool all the components in the system.
Waterluvian 373 days ago [-]
As a kid it felt so weird to me that SLR lenses could be so massive that you mount the camera to it rather than the other way around. This feels like that.
hlandau 372 days ago [-]
The retention mechanism in the PCIe spec he's quoting here isn't metal; it's literally just an extra piece of plastic on the PCIe slot. I've used motherboards with them, they do exist. You have to move it out of the way if you want to remove the card. I don't know that it really adds much mechanically though, it's just a retention mechanism. Since it's part of the PCIe slot, any load it bears is pulling on the same thing: the motherboard. Image: https://www.vortez.net/index.php?ct=articles&action=file&id=...

It does feel like GPUs are getting rather ridiculous and pushing the limits. PCIe SIG seems to keep having to specify editions of the PCIe add-in-card electromechanical spec authorising higher power draws, and sometimes it seems like these limits imposed by the standard are just ignored.

puyoxyz 373 days ago [-]
450 Watts?! Support sticks ?!?! They’ve actually gone insane. What the hell.
Thorentis 373 days ago [-]
Why do we need to keep plugging the GPU directly into the board? Why can't GPU makers ship cables that go into the PCIe slots, and then connect to the GPU, then we can mount the GPU somewhere else in the case (perhaps case makers can start adding GPU slots where the DVD drives used to go or something).
numpad0 373 days ago [-]
One of the issues with that second bracket is there are not many cases that supports it. Mac Pro, HP workstations and desktop servers, and old, cheap looking, bent metal “DOS/V” cases has it, but modern “EATX” cases often don’t. It also cannot support too much weight.

Indeed, PCI standards were for adding interfaces to personal desktop computers after all. It does seem ill suited to host 450W cooperative sub-computers.

A more common approach to heavy expansion card is VME style chassis design. Off top of my head, NeXTcube, NEC C Bus, Eurocard uses this arrangement in consumer space, and many blade servers enclosures, carrier grade routers, and military digital equipments employ similar designs as well.

voidfunc 373 days ago [-]
I predict the GPU will need to be externalized in the future with its own cooling and power system. You'll plug into the motherboard via some cable interconnect.

They're simply getting too big, power hungry and hot to keep in colocated in the case.

qwerty456127 373 days ago [-]
> Maybe we can make motherboards with a GPU slot next to the CPU slot and have a unified massive radiator sitting on top of them

Sounds reasonable, we already used to have separate CPU and FPU sockets in the distant past.

However, isn't it nice every extension card incl. GPU cards uses the same unified connector standard and can by replaced with anything very different in place? Wouldn't switching to the an MXM form-factor, introducing an extra kind of slot, be a step back? Haven't we once ditched a dedicated GPU card slot (AGP) in favour of unification already?

jxramos 373 days ago [-]
Hearing about the EVGA news recently I was asking at work trying to understand the difference between GPU chips and cards and came to understand the form factor a bit more. We were just talking about motherboard CPU sockets and how it’s easier to differentiate the CPU from the motherboard because they’re two separable components. With GPUs they’re bonded to their cards so the visual separation is a lot harder to comprehend without understanding the form factor a bit more closely. It’ll be cool to see GPU sockets on motherboards if that becomes a thing.
rascul 373 days ago [-]
Why can't I find a recent GPU that doesn't take up half my case? Do the GPU manufacturers not care about people who want more than onboard graphics but don't need the most powerful gaming GPU?
kenhwang 373 days ago [-]
You can? Both the RTX 3050 and RX 6400 come in single fan form factors not much longer than a PCIE slot. Anything weaker than that is getting beaten by AGUs.
rascul 373 days ago [-]
For reference, there used to be low profile cards available barely longer than an AGP slot. For example, here's a low profile Nvidia Gefore4 MX440 (I used to have one of these):


> Anything weaker than that is getting beaten by AGUs.

Is AGU the GPU on the CPU thing? Is it not possible (or maybe not profitable) to put something a little more performant than an AGU but doesn't need such a massive fan on a full size multiple slot card?

goosedragons 372 days ago [-]
That's basically what the Radeon 6400 can be. Here's the low profile version:


Barely longer than the PCI-E slot and only needs a single slot.

The Radeon RX 6400 is basically AMD's high end integrated GPU on a card being very similar to the Radeon 680M in Ryzen 6000 chips. But I think the dedicated memory offers an improvement over the APU solution.

kenhwang 373 days ago [-]
APU*, for accelerated processing unit. Essentially a CPU with integrated graphics, but the graphics cores can also be used for traditional CPU computing if it's not being used for graphics.

They can't put anything bigger because they run into heat, power, and memory limits.

Narishma 373 days ago [-]
Those low-end GPUs have been an afterthought for many years now. The last one I remember being somewhat decent was GT 1030 from 2016.
zekica 372 days ago [-]
RX6400 is a lot better than 1030, and is available as low profile single slot cards.
dragontamer 373 days ago [-]
What's wrong with the NVidia 1630 or 1660, or AMD 6400, or Intel Arc 380?
rascul 373 days ago [-]
Dual slot, massive fans, and only the Arc isn't extra long, from the pictures I see. I guess we have different ideas of big.

Edit: Perhaps I use the term incorrectly, I mean they take up two case slots even if only a single pcie slot.

dragontamer 373 days ago [-]

Single slot, half-profile, small fan, shorter than ATX / mATX?

This is a mini-card that'd fit in an ITX case (170mm)...

skirmish 373 days ago [-]
Simple: GPU manufacturers would much rather sell you a flagship at $1000+ than multiple basic cards at $150. Duopoly in the market and no competition, why would they reduce their margins?
Sohcahtoa82 372 days ago [-]
> flagship at $1000+ than multiple basic cards at $150

Those aren't even the same market segment.

The type of person buying a flagship card would not even be considering a basic card and vice-versa.

Also, a part of me used to get frustrated at the existence of the "basic" cards, like the GT 1030. I've seen more than one person wanting to build a gaming PC and see a budget card like that and think they're getting a current-gen card without knowing that the budget cards are horrendously underpowered. For example, the GT 1030 is about the speed of a GTX 470, a mid-grade card from 7 years before it.

Maybe some people are fine getting that weak of a card, but if that's all you need, I'd question if you even need a discrete GPU.

phkahler 372 days ago [-]
Just replace the motherboard with a GPU board with cooling solution on the bottom, and CPU socket on top along with all other connectors. Make them in 2 standard heights so cases can be standardized. The boards will have NO slots since the board IS the video card. This is of course only slightly serious, I prefer to use AMD chips with integrated GPUs and SFF. But for those that care about GPU capability lets just change the priority of the components and build the PC on top of the graphics solution.
frostburg 373 days ago [-]
This specific problem can be solved by rotating motherboard 90 degrees (there are a few Silverstone cases that are laid out like this, they also tend to have excellent air cooling performance).
jrootabega 372 days ago [-]
Yeah, I've had a Silverstone Raven for 10+ years, so I'm not sure why those types of cases haven't become the standard. They have their flaws, like it being awkward to route your cables into the top area, and needing good airspace underneath the case. But it's balanced out by being able to easily plug stuff into the top when you don't care about pretty cable routing, a focus on powerful, adjustable airflow, and less need for space on the sides. I'm guessing vertical space is the main problem for people.
seiferteric 373 days ago [-]
Put an edge mount PCIE connector on the motherboard and allow the graphics card to plug in parallel to the mother board and mount the card on the case itself like the motherboard.
rektide 373 days ago [-]
I wonder whether pcie cabling (like oculink) will scale/work with newer pcie specs.

I have long thought the bitcoin miners were onto something, with pcie risers galore. In my head I know pcb is cheap and connectors - cables arent but it always seemed so tempting anyways; very small boards, cpu & memory (or onchip memory) & vrm, and then just pipes to peripherals & network (and with specs like CXL 3.0, we kind of could be getting both at once).

pcdoodle 372 days ago [-]
This should be fixed by case manufactures working with the other players. Pick a reasonable max length and be done with it. GPUs can come with an extender to meet the case front captive bracket.

Nobody agrees on anything anymore. We need standards like those created 30 years ago. But everyone wants to do their own thing without realizing the reason for the inter compatibility is because people got over themselves and worked together.

chubs 373 days ago [-]
What if a 4-slot-wide GPU had 3 'dummy' slots that plug into the other unused PCI slots, no electrical connection, and only acted as support?
fredoralive 372 days ago [-]
Modern motherboards often don't have PCIe slots next to the x16 "GPU" slots, as they'd nearly always be under the cooler, so why waste PCIe lanes (and BOM cost) on a slot there? The increasing trend towards x3 and x4 width cards might mean you'd be able to use the next but one slot (assuming they don't get removed soon...), but even then there isn't really standardisation about what type of PCIe slot goes (or doesn't get fitted) in which position on the motherboard.
unethical_ban 373 days ago [-]
No, we don't. Show me the articles where peoples' PCIe slots are ripping out.

What we could do is have AIO cooling like CPUs, more affordable than the current solutions or the "water block" designs from the brands.

Or, have more sandwich designs like Sliger which have a mini itx and a PCIe card parallel and connected via a ribbon. I don't think there is any claimed performance loss due to the cable.

TT-392 373 days ago [-]
I think a start would be providing an extra screw hole in a standardized place, so that case manifacturers can design for this, instead of putting the card between 2 plastic clamps. A problem here would be length differences in cards though. But I think that it being in the middle of most long cards, and at the end of the smallest card isn't the craziest idea.
sp332 373 days ago [-]
I suppose it's possible, but I have yet to see an actual GPU that couldn't be fixed by properly supporting the back bracket, as shown in this video. https://www.youtube.com/watch?v=liChy76nQJ4&t=591 (Starts around 10:00 in)
JonChesterfield 373 days ago [-]
We don't _need_ new motherboards. Just stop standing the things up on their side. Horizontal case, no problems.
lstodd 373 days ago [-]
All you have to do is put an aluminum I-bar on top of the card if the cooler itself doesn't provide adequate rigidity, which I doubt.

I'd guess if excessive stress on the PCIe slot was a problem, it'd be solved by combining a good 2-3 slot mount on the back side and enough aluminium+plastic to hold the rest.

bigmattystyles 373 days ago [-]
Can you have a cable extended pcie socket or would that introduce too much latency? You could argue we just need a new form factor. And if I can put the mobo parallel to the GPUs, problem kinda solved. Or just go back to the desktop form factor like someone else said, remove torque from the socket.
anigbrowl 373 days ago [-]
What if we just have blocks of quartz and use laser arrays to build photonic switching junctions, no more cooling problems because it's just photons ¯\(°_o)/¯

Seriously though, I imagine it's only a matter of time before these engineering decisions are themselves handed off to machines.

373 days ago [-]
andrewmcwatters 373 days ago [-]
Problem is that the card retainers in the spec would all, I suppose, need to align with the chassis. Card widths are highly variable, so all manufacturers would need to change their card designs to allow for at least providing aftermarket retainers on them.
rasz 372 days ago [-]
For GPUs to Collapse Under Their Own Gravity they would need to go below https://en.wikipedia.org/wiki/Schwarzschild_radius
zargon 373 days ago [-]
1. I don't see the problem with GPU support sticks. I used one at least 12 years ago.

2. The real problem, in my opinion, is out of control power consumption. Get these cards back down to 200 W. That's already twice the power delivery of a mainstream CPU socket.

m463 373 days ago [-]
> The real problem, in my opinion, is out of control power consumption

That is the design spec for laptops, but not many people game on laptops because although they are more efficient, people would rather go 2x or faster and use more power

arvinsim 373 days ago [-]
We need more cases for horizontally oriented motherboards.

I was also thinking of a case where it can handle the cooling of a deshrouded GPU. Perhaps we should delegate the cooling options to the user without having to void warranty.

jwlake 373 days ago [-]
Why doesn't the "video card" just come with a processor socket?
draluy 373 days ago [-]
Just mount the Graphics card vertically using a raiser cable. It more of a PC case problem that a motherboard problem. Also, it allows for better form factors, with spread out components and bigger fans.
nottorp 372 days ago [-]
We need a concerted effort to reduce chip power consumption.

Hopefully the energy prices in Europe will force chip makers to work on that. I mean, only if they want to sell something over here.

fancyfredbot 373 days ago [-]
NVIDIA makes a socketed version of their data center GPUS. The socket is called SMX. It would be cool if consumer board partners and motherboard manufacturers used it too.
hn_zorba 373 days ago [-]
From now on, we mount the motherboard on the GPU.
rybosworld 373 days ago [-]
I always thought it interesting how most tech companies prioritize shrinking the hardware but GPU's seem to be an exception.
fjfbsufhdvfy 373 days ago [-]
The technology is getting smaller! It's just that at the same time, people still demand orders of magnitude higher performance. There are physical limits to how small a cooler can be, and that takes up like 95% of the space.
e1ghtSpace 373 days ago [-]
Nobody wants a small gpu.
doubled112 373 days ago [-]
How small are we talking?

I really like the Mini ITX form factor and have multiple PCs that fit on IKEA shelves.

I prefer shorter/smaller GPUs. Better airflow. Easier at build time too.

My EVGA RTX 3050 is about the right balance.

wengo314 373 days ago [-]
why not simply use a case where the board is placed horizontally, that should not damage the slots?

or provide a dedicated gpu slot with a riser-like mount that allows for gpu to be mounted separately from the actual board ( something what laptop owners do with external gpus) ?

this way gpu could be any size and might have cooling on either side - or an external solution.

hnaccountme 373 days ago [-]
Maybe its time to have a separate socket for the GPU, like how CPUs are installed. Might need to have gddr slots too. LOL
htrp 373 days ago [-]
Lets just attach the rest of the components to the GPU and screw the gpu directly into the case.....
rawoke083600 373 days ago [-]
Lol it's a matter of time, until you buy a GPU and "just" add a CPU and some RAM.
lukaszkups 373 days ago [-]
To the author: I see what you did there and I like that Cyberpunk: Edgerunners reference ^,^
everyone 373 days ago [-]
What always annoys me is that with air cooled gpu the hot air goes down.
flenserboy 373 days ago [-]
Perhaps game makers could focus on gameplay and story instead of whatever unnecessary detail is chewing through so much data. The big iron is great for actual work, but is pure overkill to have in some junior high kid's room. Just an idea.
birdyrooster 373 days ago [-]
Have you seen cases these days? They already account for this.
ThrowawayTestr 373 days ago [-]
Are the screws into the case not enough?
ruffrey 373 days ago [-]
anigbrowl 373 days ago [-]
Explicit instructions considered harmful
bfgoodrich 372 days ago [-]
de6u99er 373 days ago [-]
I don't agree. I think we need new GPU's with wider instruction sets.

Or maybe FPGA's onboard for custom per use case. I hope that's why AMD merg d with Xilinx.

Victerius 373 days ago [-]
With a power consumption of that magnitude, no wonder power grids are buckling everywhere. People's hunger for power is seemingly insatiable.
db48x 373 days ago [-]
According to the Steam Hardware Survey (which you can view for yourself at <https://store.steampowered.com/hwsurvey/Steam-Hardware-Softw...>), the RTX 3090 makes up less than 0.48% of all video cards in use by gamers. The 450W power figure is for the new RTX 4090, which will no doubt take a while to reach even that level.

It might be nice to have more information than that, but it honestly doesn’t seem like a huge problem. Your fridge probably costs you more over a year than even a 4090. See also air conditioning.

CyberDildonics 373 days ago [-]
Power grids are not buckling everywhere. Also this card uses less than a third of a single space heater. Finally, this card has not been released yet, so it isn't causing anything.
Sohcahtoa82 372 days ago [-]
The power grid is only crumbling in places where political corruption is blocking fixing the grid.

See Texas.