And a standard for modular plugs on PSUs, so you have the same pinout for every PSU, making aftermarket shortened cables way easier. That would be the dream.
-
Save 15% on ALL SFF Network merch, until Dec 31st! Use code SFF2024 at checkout. Click here!
Motherboard SFF-Forum's Design a board thread
- Thread starter confusis
- Start date
While there are pitfalls with the approach, a consumer run standards/advocacy organisation for Small Form Factor systems might be one route to getting manufacturers on board with SFF designs. Things like shorthand designations for GPU card sizes, alternatives to ATX12V/EPS12V (would definitely need industry assistance there), possibly even new board form factors, will only fly for consumer products if multiple manufacturers can be convinced that cooperating is more worthwhile than everyone doing their own thing for product differentiation.
I don't know which other manufacturer you are thinking of, but ASUS, ASRock, MSI, ECS have been selling LGA1150 Thin mini-ITX boards. MSI has the least number of Thin Mini-ITX boards in their catalog but they also made interesting MoDT ITX boards in the past too.
Certainly true, but this does happen though. Look at SFX-L how Silverstone pushed the standard onto the OEM, allowing other brands to adopt it. Or even earlier, how they focused on SFX. While this wasn't a sure thing for a long time, it's still possible with the right company, momentum and consumers.
I like this idea a lot. Getting rid of the 24 pin connector would be a huge benefit in smaller cases. Thin ITX already does something like this by taking a single voltage and generating the other needed voltages on-board. They just use a barrel connector or 2-pin though, while EPS12V has a theoretical maximum of 338W, which is more than enough power for pretty much any mini-ITX board/CPU and fully-powered PCIe slot (75W).
It's a minor side point, but I can't help but to set the record straight when this comes up: it was Sirfa/High Power that developed the first SFX-L PSU after we contacted them about it. The specs we gave them were virtually identical to what eventually came to market: 130mm depth, 120mm slim fan, fully modular, all or most of the rated power available @12v.
Of course, it still wouldn't have happened if Sirfa hadn't developed and shown a prototype at Computex, and Silverstone hadn't then made the first production order for what became the SX500-LG. It was also Silverstone that came up with the name 'SFX-L' for the form factor.
Speaking of SFX-L (and sorry for going a little off topic), I have in my hands one of the new Sirfa 700W units. It's a preproduction model which came to me from Dirac, so it has no labeling indicating the model or specs, and looks otherwise identical to the 500W unit. I've only done some limited testing with it so far installed in one of my M1s with a 3770K (stock) and GTX 780, and I've not noticed any irregular fan noises from it. The fan, which appears to be the same model as the 500W, is completely off during idle, and only spins up during gaming, where of course it's completely drowned out by the reference 780. It also shuts off again almost immediately on return to idle, which I thought was interesting.
Whoops, I meant three. I only know of four Gigabyte boards, two ASUS boards and one from ASrock. Haben't seen MSI or ECS, though. Gigabytes boards also support 12V input already, so there's that
An additional benefit of generating voltages on the board is that you can make the 12V rail in the PSU stronger, so if the 3.3V and 5V outputs aren't used, you don't waste any potential power. At least I like to think it works that way.
Yeah maybe we'd just need a 4pin connector for the PSU-mainboard logical interconnect, so PS_ON, PWR_OK, 5VSB and ground and have the 12V being input by the 4pin or 8pin ATX12V connectors. Maybe add a 2pin version to save space in certain scenarios.
The problem is: Silverstone can do that. We would have to invest a lot of work to actually sustain this for long enough so out ideas can catch on in the retail market.
Very interesting indeed. So it seems like the new design really does have a proper fan controller.
Credit goes to iFreilicht because the idea wasn't mine for the 6 or 8-pin plug, but I support his solutions as being viable. I'm not a supporter of putting more PSU components on the motherboard though, it makes them much more important to be "perfect" (coil whine and voltage ripple for example).
Ah yes, I forgot about that, sorry I said it like that. Credit where credit is due.
Very interesting. Here's hoping they perfect the issues so Silverstone can't screw it up with their recent fanless solutions.
I've also not heard strange noises with my 500W SFX-L unit, although I can when I would live in a music recording studio like envireonment maybe, or put my ear on the PSU. Too faint for normal use.
A few questions/comments:
1) Has anybody codified the SFX-L form factor as a "standard"?
2) Is the "Power Supply Design Guide for Desktop Platform Form Factors" (available from formfactors.org) simply Intel making a proclamation from on high (or by default), or is there an industry body that develops and maintains these "standards" and Intel merely published the guidance?
3) It would seem likely that if anybody could reduce the size of the standard 24-pin PSU connector, it would be the motherboard manufacturers. They would probably be overjoyed to free up half the connector space as well as all the traces in order to put more "cool stuff" on the mobos, especially for small (mATX and smaller) mobos.
4) Assuming 3) gets done, PSU vendors would likely offer adapters from 24-pins to whatever the new count is that would connect to the business end of non-modular mobo cables and/or additional new modular cables entirely.
5) I don't think you'd get the pinouts on the PSUs themselves to change very quickly, unless the aftermarket PSU modular cable market gets hotter.
1) Has anybody codified the SFX-L form factor as a "standard"?
2) Is the "Power Supply Design Guide for Desktop Platform Form Factors" (available from formfactors.org) simply Intel making a proclamation from on high (or by default), or is there an industry body that develops and maintains these "standards" and Intel merely published the guidance?
3) It would seem likely that if anybody could reduce the size of the standard 24-pin PSU connector, it would be the motherboard manufacturers. They would probably be overjoyed to free up half the connector space as well as all the traces in order to put more "cool stuff" on the mobos, especially for small (mATX and smaller) mobos.
4) Assuming 3) gets done, PSU vendors would likely offer adapters from 24-pins to whatever the new count is that would connect to the business end of non-modular mobo cables and/or additional new modular cables entirely.
5) I don't think you'd get the pinouts on the PSUs themselves to change very quickly, unless the aftermarket PSU modular cable market gets hotter.
1) Don't think so.
2) Good question, I'd like to know that as well. I think that document is a compilation of all Form Factors currently in use, ATX has its own standard. There are groups that do standardise stuff like this, like VESA or PCISIG, but they have a lot of member companies that all work on those standards together.
3+4) If they do that, then Motherboard manufactures would have to offer adapters first, not the PSU guys. But yeah I'd like to see them do it.
Interesting discussion. It seems to me that the whole ATX power supply spec is a dinosaur of a legacy spec that needs to be trashed asap. But that's not what I want to dicuss.
I have always thought it would be interesting to see a motherboard into which you had two sockets, one for the CPU and one for the GPU. This would get rid of all the crap around GPU add-in boards. It would make powering and cooling the GPU a lot easier. I have raised this idea in the past and people have told me about the reasons why this is difficult - people want more flexibility with GPUs, ability to upgrade/swap/change manufacturer - which would be difficult with a socketed GPU. And of course, as discussed in this thread, space on motherboard is at a premium, with all the additional circuitry required for the GPU. I'm sure there are other reasons as well.
BUT - now with AMD's new HBM chips which have memory built in to the GPU package, this would save a LOT of board space and get rid of any concerns or having the correct memory on the board for the GPU. Upgrades and swaps would be easier and potentially you could have chips from different vendors fitting the same socket (pigs might fly). Would it be possible to make a motherboard with a HBM CPU socket?
I have always thought it would be interesting to see a motherboard into which you had two sockets, one for the CPU and one for the GPU. This would get rid of all the crap around GPU add-in boards. It would make powering and cooling the GPU a lot easier. I have raised this idea in the past and people have told me about the reasons why this is difficult - people want more flexibility with GPUs, ability to upgrade/swap/change manufacturer - which would be difficult with a socketed GPU. And of course, as discussed in this thread, space on motherboard is at a premium, with all the additional circuitry required for the GPU. I'm sure there are other reasons as well.
BUT - now with AMD's new HBM chips which have memory built in to the GPU package, this would save a LOT of board space and get rid of any concerns or having the correct memory on the board for the GPU. Upgrades and swaps would be easier and potentially you could have chips from different vendors fitting the same socket (pigs might fly). Would it be possible to make a motherboard with a HBM CPU socket?
You'd have to specify a Socket for that, of course, but in theory it could maybe work. The problem I see is that you maybe can't even upgrade your GPU even if it was by the same manufacturer, because later products could need different VRMs and stuff like that. You never know what the next generation brings, a socket is too limiting in what can be done as of now.
If something like this is meant to fly, you'd have to standardise a completely now interconnect both physically and electronically that accomplishes the same PCIex16 now does, and I think there is very little incentive to do that as it can only be applied to a very specific use-case and doesn't offer any additional benefits. You'd still need a large cooler to cool the GPU, so there is little space to be saved. You'd need special boards if you wanted to use co-processors or multi-GPU configurations.
PCIe is a one-fits-all connection that is extremely fast and versatile and I don't think anyone really thinks about leaving it for something different.
If something like this is meant to fly, you'd have to standardise a completely now interconnect both physically and electronically that accomplishes the same PCIex16 now does, and I think there is very little incentive to do that as it can only be applied to a very specific use-case and doesn't offer any additional benefits. You'd still need a large cooler to cool the GPU, so there is little space to be saved. You'd need special boards if you wanted to use co-processors or multi-GPU configurations.
PCIe is a one-fits-all connection that is extremely fast and versatile and I don't think anyone really thinks about leaving it for something different.
No, it is intended as a CPU-GPU link as well. What isn't clear yet is if it will only do so for HPC (where throwing in custom silicon is not a problem), or if Nvidia will either be hashing out the legal details to release their first intel chipset after 7 years or persuading Intel (and possibly AMD) to integrate an NVlink controller on die.
Personally, I think NVlink is going to end up confined to HPC entirely, with GPU-GPU links for consumer cards remaining with the SLI bridge or moving entirely to PCI-E (as Crossfire has done). Having two entirely different high speed bus controllers on a die is going to eat up way too much area for consumer margins.
Where does it say that? It just says it enables fast CPU-GPU communication, which it apparently does by moving inter-GPU traffic to the new interconnect and distributing the loading process of large data over multiple GPUs, but it doesn't say that it actually is an interconnect between CPU and GPU. Would be very strange if it was because then Intel would have to integrate it into the CPUs in order to make it worthwhile. If they didn't, PCIe would still be bottlenecking it, right?
But yeah I agree, I don't see this coming to the consumer market. Would be nice to see bridgeless SLI, though.
It's hard to find info on this part of the technology but it also covers a new form factor with mezzanine type connectors: http://cdn.wccftech.com/wp-content/uploads/2014/03/NVIDIA-Pascal-GPU-Chip-Module.jpg
Personally I would be very skeptical on Nvidia implementing connectors on common hardware. They have always been very greedy in regards to industry-wide tech they've researched, so I wouldn't be amazed they would ask hefty royalties for using these.
In this presentation. NVLink between Power and ARM CPUs and GPUs is shown explicitly, but all slides involving x86 CPUs show a PCI-E bus between CPU and GPU implying NVLink may remain as inter-GPU only for consumer cards (if it even makes it to consumer cards).
Similar threads
