Asus details its cable-free GC-HPWR connector for 600W gaming GPUs
videocardz.com
videocardz.com
This also means you can deal with whatever bespoke voltages you want for the power delivery you need, as opposed to being constrained by standards for the system.
Been there, done that:
The external power you're talking about would be like 1 volt at hundreds of amps. What's the point of that? The connector and cable would be huge, think copper feeders. The existing set up where a higher voltage (12v) is stepped down to all dynamic target voltages (Vcore, memory etc) is fine and there isn't anything wrong with that. No need to reinvent the wheel. I don't want 1 external power supply per graphics card, what is the point of a modular internal power supply then?
The rest of the PC can run off, what, 200 watts, and fit on a nano motherboard with basic cooling? Then comes the big ass space heater GPU that draws most of the power and disrupts airflow for everything else. Why not have it in its own little cage, almost like a game console? It can get better cooling that way, its own power supply that maximizes output instead of efficiency, and easy upgrades. Connect it when you wanna game or do AI stuff, leave it completely off the rest of the time.
If it had its own little web enabled controller, maybe you could even rent it out for processing workloads in the clouds.
I always saw them as intended for laptops rather than desktops - are there any downsides or performance penalties compared to slotting directly into a PCIe slot?
Latency, for one. All signals must at least pass through two Thunderbolt/USB4 chipsets on their way to and from the GPU, and at least for anything but high profile AMD CPUs Thunderbolt is connected to the southbridge instead of northbridge (where the x16 GPU slot is).
That's too bad, if so :(
(Not a hardware guy. Didn't realize there physical/electrical limits to these.)
I've been building PCs since I was a kid, and I'd still rather gouge my eyes out with a rusty spoon than have to change a PC power supply - it's just so, so damned fiddly and time consuming in almost every case I've ever seen :(
Where we used to fit all the components from the same side of the case, modern cases have the PSU fitted from the reverse side, allowing the cabling to be concealed yet reasonably easy to access. I believe this was intended to make things pretty for the glass-window enthusiast crowd, but it also makes changing the power supply easier.
There are also substantially fewer components in general. Things like networking and sound have been on the motherboard for decades, of course - but now 5.25-inch drive bays are also gone, with 3.5-inch bays in decline. And of course with M.2 you might not need any other drives.
All of which means there are fewer power cables to plug in than ever before!
Re: voltages. I mean the opposite. The idea here is that we could supply, say, 50V for the graphics card to step down now that we're no longer constrained by system voltages. Now there is much less current going through the fiddly connector. A 2200W power supply constrained by 12V is delivering at least 180A.
What hurts 48V adoption is typically component cost. In the past there was a large price jump between 12V and 48V-60V capable circuits, with a lot of competition and demand at 24V from industrial stuff but less at 48V. Part of the problem is the huge ratios of 10x-50x between the voltages and the idea that the semiconductor switching elements have to be able to handle the peak voltages encountered in the system. In order to realize a 48V system at minimum cost you kind of need a flyback style power converter, leaning on a transformer's winding ratio to help. Dealing with custom magnetics sucks and engineers have in the past related to me how much they try to avoid dealing with magnetics. The frequency you want to use is high and this makes the magnetics smaller but increases the stress on the switching element (MOSFETs). The power you get off a flyback-buck like this can be a little noisy if not postregulated so you might need clever tricks to regulate it better to reduce that in the first place. I have noticed that high powered PoE adoption and lithium battery pack demand (like for electric bikes etc) is driving some of the costs on these things down a bit.
For starters - a typical 15amp outlet is only supposed to have a 1,440W continuous load, 1,800W peak.
240v is even less common.
So ... yeah, you can do it, but you're talking about the equivalent work to having an electric car charger put in your office.
Luxury! Where I live the typical socket is only 10A, and the most common standard voltage is 127V (it varies by city, some places have 220V), so the maximum would be around 1,2kW.
(There's also a 20A socket, with slightly larger pins, but it's mostly found on sockets dedicated for window air conditioners.)
Allow me to introduce you to the 2005 era "ASUS Extreme N7800GT Dual" [1]
Power is supplied to the board via an external PSU, rated to provide 80W of output power from 12V DC. You can power the board from your PSU, provided it's ATX2.0 and has a suitably meaty current ability [...] obviously the external PSU is the recommended method of keeping the board fed with the power it needs.
(The product actually combined two GPUs, linked with SLI, on a single board.
With the absurd price of over £600 the press considered it a very very expensive boutique piece of hardware.
[1] https://www.hexus.net/tech/reviews/graphics/3654-asus-extrem...
The last Intel Mac Pro had an extremely similar power delivery method, and it really makes for a very clean look + simplifies things ever so slightly when replacing cards.
I’m sure people will say that neither of those things are worth it for the change involved, but I’m still hopeful that this becomes part of the ATX standard going forward
Also, with this connector being so far away from the mounting brace to the case, would GPU sagging torque/bend/ this connector and all 600w of power going to it? A 600w GPU would have a big 'ol cooler and have big sag, right?