Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

We Need New Motherboards Before GPUs Collapse Under Their Own Gravity

erikmcclure.com

Post reply on HN
Print view View original

351–360 of 362 posts

Re: We Need New Motherboards Before GPUs Collapse Under Their Own Gravity

#351

Post by gh02t »

Sohcahtoa82 wrote: post #343
gh02t wrote: post #118

Earlier quoted context omitted.

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.

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 en…

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.

Re: We Need New Motherboards Before GPUs Collapse Under Their Own Gravity

#352

Post by Taniwha »

userbinator wrote: post #316
Taniwha wrote: post #282

Earlier quoted context omitted.

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 gr…

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.

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)

Re: We Need New Motherboards Before GPUs Collapse Under Their Own Gravity

#353

Post by Taniwha »

im_down_w_otp wrote: post #321
Taniwha wrote: post #211

Earlier quoted context omitted.

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)

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.

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

Re: We Need New Motherboards Before GPUs Collapse Under Their Own Gravity

#354

Post by chickenimprint »

qwerty456127 wrote: post #271

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…

What exactly do you want to expand it with though? Sound cards are dead and everything besides accelerators is USB now.

Re: We Need New Motherboards Before GPUs Collapse Under Their Own Gravity

#355

Post by DaiPlusPlus »

imtringued wrote: post #326
rocket_surgeron wrote: post #193

Earlier quoted context omitted.

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 t…

Honestly just use chiplets technology for more cores and get rid of the multi GPU concept entirely.

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.

Re: We Need New Motherboards Before GPUs Collapse Under Their Own Gravity

#356

Post by kuschku »

rocket_surgeron wrote: post #331
kuschku wrote: post #288

Earlier quoted context omitted.

> 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)

No. 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…

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.

Re: We Need New Motherboards Before GPUs Collapse Under Their Own Gravity

#357

Post by 5d8767c68926 »

nextaccountic wrote: post #173
greggsy wrote: post #166

Earlier quoted context omitted.

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 thes…

> 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?

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/

Re: We Need New Motherboards Before GPUs Collapse Under Their Own Gravity

#358

Post by 7speter »

chii wrote: post #192
7speter wrote: post #189

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 ho…

> 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.

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.

Re: We Need New Motherboards Before GPUs Collapse Under Their Own Gravity

#359

Post by bofh23 »

somat wrote: post #293
0x38B wrote: post #207

Earlier quoted context omitted.

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...

The sgi onyx reality engine "graphics card" was composed of like 5 motherboard sized pcb's. https://hardware.majix.org/computers/sgi.onyx/onyx.shtml

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.

Re: We Need New Motherboards Before GPUs Collapse Under Their Own Gravity

#360

Post by bofh23 »

arvinsim wrote: post #315
zaptheimpaler wrote: post #36

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.

Isn't this what the thunderbolt enclosures with GPUs are trying to do?

I’m pretty sure Thunderbolt only has the equivalent of 4 lanes of PCIe 3.0 while GPU cards tend to be 16 lane.
Post reply on HN