I've been saying this for a while now, but the ATX standard needs a complete modern redesign centered around the GPU, instead of the CPU like in the 90's, when that component was the most important and energy intensive one.
But since Intel is the original designer and contribuitor of the ATX standard, I doubt they're looking forward to give up the throne in the system to Nvidia/AMD till they gain any majority share in the GPU market which won't happen in my lifetime.
The ATX standard is far too long in the tooth already and you can see various people doing their own hack jobs at home, while case and motherboard manufacturers try to innovate around it without Intel, going in their own directions with non-standard deviations and hacks, meaning components are expensive due to lack of economies of scale and cross-compatibility is a mess.
I guess the lack of interest in modernizing the ATX standard also stems form the fact that PC sales have been on the decline for over a decade(pandemic excluded) due to the rise of mobile use, with tower PCs seeing an even bigger decline as most people and companies who still need PCs use laptops now, meaning ATX style PCs have made a 360 turn back to being a rarity for hackers/geeks/gamers instead of something everyone had to have in their homes and offices.
But what I personally want today is not another quirky square box occupying space on my desk/floor, but a stealth desk PC like this one (watch the video it's worth it): https://www.youtube.com/watch?v=UQ6ueTe8KQM
If ATX is being redesigned today, than it will not become better but worse for end users. Things like moving major power supply to the motherboard - for instance like in the 12VHPo standard.
Not to mention that most of the power provided by modern PSUs come from the 12V rails. The 3.3 and 5V rails are almost vestigial at this point, no hardware actually runs directly at those voltages, so removing them would not have any negative effect.
Because it moves complex circuity to an already very complex and expensive PCB, making it even more expensive. A power supply is likely to outlast multiple (re)builds of a desktop computer; a motherboard is likely not - especially not Intel motherboards.
I worked on 1 or 2 boxes designed like that, and they were just annoying. Instead or a "redesign" that I agree with you we need; Simply described, it's an upside down ATX.
OH yeah I remember BTX, but that design was dated even for that time since it was only meant to stop Pentium 4s from overheating, not accommodate the monster GPUs of today. It was clever that it would Cool the PSU and CPU with one fan but it was still overall crap.
Also, just because a poor replacement suggestion failed once in the past, doesn't eman it's an excuse to give up on replacing ATX with something better.
Rage converted, yes. Looks like the cpu is air cooled on top of two other computers, an ups, and maybe a third or more accessory computers. All inside a bathroom with probably poor ventilation. None of my bathrooms have return air. Maybe too many heat sources in too small of a space.
A server guy I follow put in a mini-split into his server room to handle the extra load of ~ $100/mo electricity worth of server hardware.
I water cooled my cpu and considered a second loop for gpu but found it was un-necessary according to some research. My new problem is airflow to the office due to heat and no return air in that room either. I run a vornado fan and propped the door open a specific amount where it wants to suck the door closed. It seems to work for now but I’m also looking into a mini-split or air duct booster as my office is the hottest room of the place.
also looks like that case would be starved for air at the best of times. I wonder how much of this effort could have avoided if they had simply ordered a case with better airflow off Amazon and returned it if testing didn't show an improvement.
I suspect that longer PCIe extender cables would have signal quality issues (so far with my setup I haven't seen any issues, mem copy throughput is as expected and no PCIe AER errors reported)
I wonder what applications he has in mind. Maybe GPGPU/ML training kind of stuff? Rampant speculation:
Cable extenders are known for messing with signal integrity. Flipping bits over long wires takes a while. ML training seems (I could definitely be wrong here) less latency sensitive than classic GPU applications, like gaming. But it does want the bandwidth. Could there be room for an active cable with repeaters? Does anyone make a cable with, like, little flip-flops in it? Can an extender act as a little buffer or pipeline or something like that?
The PCI AER (advanced error reporting) can detect issues quite well & report them. The author mentiona that they sort of expected problems, but no, AER hasn't reported any.
Links have to be very smart & capable of self training in modern PCIe to perform at the speeds they do: they have link equalization. So there's incredible diagnostics builtin.
As for active cables, Marvell just announced Alpine P, which is a PCIe6 retimer that capable of working in active electrical cables (and also active optical cables!). I don't know if they buffer data or not - if so it would be a small amount of data I'd expect - but they do receive & retransmit. https://www.servethehome.com/marvell-extending-pcie-gen6-rea...
Just random LLM use (no training) and some cuDF experiments so far. So far I haven't noticed any issues (and I tend to investigate any unexplained glitches).
I should have noticed the url/username match, haha.
Neat project! I’ve done a couple (basic, consumer, no modifications of the case) mini itx builds and always wished I could get the GPU out of the case. Stick the hot thing out in the world, where it has infinite airflow, seems like the most obvious thing to do, but it really isn’t catered to well enough…
The riser cables look delicate and likely to introduce latency.
I think mainboards, particul;arly ITX, could have PCIe sockets on the back or on the edge, and/or at right-angles - for example you could mount a GPU on it's side next to the mainboard, plugged into a right-angled socket on the edge of the board.
One alternative to PCIe risers is OCuLink/SFF-8611. It lets you plug one end of a thin cable into a PCIe (or M.2) card and the other into a board with a PCIe slot.
The main downside is that it cuts bandwidth to PCIe x4, but for most usage (e.g. LLM inference), that would be fine.
That's a topic for another blog post, but I use these as servers, so don't need direct HDMI/display connectivity - but one reason is to reduce noise. And I don't have a garage!
It might be their spare/basement bathroom that they're not actually using. In the background you can also see that they're using it to store their luggage cases.
But since Intel is the original designer and contribuitor of the ATX standard, I doubt they're looking forward to give up the throne in the system to Nvidia/AMD till they gain any majority share in the GPU market which won't happen in my lifetime.
The ATX standard is far too long in the tooth already and you can see various people doing their own hack jobs at home, while case and motherboard manufacturers try to innovate around it without Intel, going in their own directions with non-standard deviations and hacks, meaning components are expensive due to lack of economies of scale and cross-compatibility is a mess.
I guess the lack of interest in modernizing the ATX standard also stems form the fact that PC sales have been on the decline for over a decade(pandemic excluded) due to the rise of mobile use, with tower PCs seeing an even bigger decline as most people and companies who still need PCs use laptops now, meaning ATX style PCs have made a 360 turn back to being a rarity for hackers/geeks/gamers instead of something everyone had to have in their homes and offices.
But what I personally want today is not another quirky square box occupying space on my desk/floor, but a stealth desk PC like this one (watch the video it's worth it): https://www.youtube.com/watch?v=UQ6ueTe8KQM