What to expect for ATX 3.0 PSUs?

[Pink Jazz]

Yes, I know this is probably a way off, but I was wondering, if/when there is an ATX 3.0 PSU specification, what changes do you expect will be made?

I expect that the -12V rail will be deprecated or even dropped entirely, since the only thing left that still uses this rail are conventional PCI slots which is the process of being phased out in favor of PCI Express and/or onboard components. The requirement for the -5V rail was dropped due to the obsolence of ISA, and is not found at all on modern PSUs.

What other changes do you think will be made? Perhaps an 18V rail could be added, which could lead to additional efficiency gains. It seems that pink wires would be the next logical color for 18V.

 

[Mr Evil]

I saw a picture of an OEM AM4 motherboard* recently that had an 8+8 pin connector instead of the usual 24 pin connector. Since it's an OEM board there is no telling what they have actually wired up to it, but it could well be just 12V. That's the direction things have been going for a while, and I expect the trend to continue.

Switching from 12V to something higher would be nice, but it would probably be better to choose 24V or 48V, since those are already used in other areas, which would make it easier to standardize on connectors etc.

* Please forgive me for linking to that site!

 

[Pink Jazz]

I wonder, besides pink, what other colors could be used for a potential future higher voltage rail? Most of the colors are now taken, with brown being the newest addition. Perhaps light blue?

While I personally would love to see my case full of snazzy pink 18v or 24v wires since pink is my favorite color, some gamers might think all the pink looks a bit too feminine.

However, anything higher than 12v would probably be gradually phased in, probably starting with the CPU and the motherboard, and eventually to the RAM and the successor of PCI Express. I would think one place to start for the motherboard would be to put a wire for that new rail in the position where the old -5v wire once was.

 

[dave_the_nerd]

I wonder, besides pink, what other colors could be used for a potential future higher voltage rail? Most of the colors are now taken, with brown being the newest addition. Perhaps light blue?

While I personally would love to see my case full of snazzy pink 18v or 24v wires since pink is my favorite color, some gamers might think all the pink looks a bit too feminine.

My understanding is that generally speaking, higher-voltage / lower amperage is more efficient and cooler. If it means they can power a 500w GPU, I think gamers would learn to love it no matter the color.

However, anything higher than 12v would probably be gradually phased in, probably starting with the CPU and the motherboard, and eventually to the RAM and the successor of PCI Express. I would think one place to start for the motherboard would be to put a wire for that new rail in the position where the old -5v wire once was.

Ack, no - if you start adding new high-voltage wires, you REALLY WANT TO create a new/different physical connector. Otherwise a lot of poor schlubs with slightly older hardware are going to run 48v to their -5v lines and fry their motherboards.

 

[Pink Jazz]

Ack, no - if you start adding new high-voltage wires, you REALLY WANT TO create a new/different physical connector. Otherwise a lot of poor schlubs with slightly older hardware are going to run 48v to their -5v lines and fry their motherboards.

I see what you mean. Perhaps it can be an another extension of the ATX12V 2.x main connector with two 24V pink wires and two common wires, making a 28-pin mainboard connector, where the extension can be separated from the 24-pin connector to allow the installation in a ATX12V 2.x system. The new 8-pin 24V CPU connector would have four pink wires and four black wires can replace the 12V connector with different notches to prevent its insertion into a 12V CPU connector and vice-versa, with new PSUs providing both connectors initially (I remember some early ATX12V 1.x PSUs providing both the 12V CPU connector and the older pre-12V Auxiliary connector).

Still, even the current ATX12V 2.x PSUs aren't recommended for pre-12V systems (especially the early AMD Athlons which were very 5V heavy) since most pre-12V systems will overload the 3.3V and 5V rails unless it is a high power unit, which still won't even be close to the advertised power since much of it is on the 12V rail. Plus if there are any ISA cards installed they won't work due to the missing -5V rail.

 

[hojnikb]

They should ditch everything other than +5V and +12V rails. Also get rid of 24pin connector and replace it with 6+2pin or something like that.

 

[Pink Jazz]

Note that some manufacturers used pink for a 5V sense wire. However, since this was a non-standard practice (unlike the standardized brown 3.3V sense wire), I am not sure if this really matters much for the color of a future higher voltage wire.

 

[MrTeal]

I saw a picture of an OEM AM4 motherboard* recently that had an 8+8 pin connector instead of the usual 24 pin connector. Since it's an OEM board there is no telling what they have actually wired up to it, but it could well be just 12V. That's the direction things have been going for a while, and I expect the trend to continue.

Switching from 12V to something higher would be nice, but it would probably be better to choose 24V or 48V, since those are already used in other areas, which would make it easier to standardize on connectors etc.

* Please forgive me for linking to that site!

If you look at that HP board, it does look like there's another couple high current DC/DC convertors in the bottom right by the PCIe slots. I don't think it's outside the realm of possibility that they're generating the relatively minor amounts of 3.3V and 5V power that a modern MB needs on-board.

The bigger problem in a general sense if that for it to be really useful, you'd need to remove the minor rails from the PSU itself and just feed out 12V. That would limit compatibility with existing hardware like storage and optical drives, fan controllers, etc.

If they do go with a higher voltage rails, I wonder if they don't skip 24V and just move up to 48V to align better with the DC power distribution you see in the server space.

 

[pitz]

-12V has been gone for ages. A supply I bought in 2005 didn't have it. I'm pretty sure PCI didn't need it either. The only real "use" for it was RS-232 serial ports, and those have have had charge-pump-based MAX232's or similar available for the past 20 years. BJT TTL logic required bipolar supplies, but computers have been nearly all unipolar CMOS for at least 20 years now.

move up to 48V to align better with the DC power distribution you see in the server space.

48V is a standard that came about on account of telecom "central offices" needing 48V for the POTS standard. The amount of that gear installed is waning, and nearly everyone has moved to 120V and even 240V distribution for datacenter-like applications.

Personally I suspect the next generation of ATX spec will define a way of allowing the PSU to communicate with the host motherboard digitally through the ATX connector. Through something resembling USB or i2C-based signaling. So that PSU vendors can incorporate voltage and current instrumentation into their PSU's. As well as read-back of PSU internal fan speeds and some host-based control capabilities of the internal PSU fans.

Going a step further, in the future, I expect PSU's to incorporate some sort of rudimentary capability of receiving signals over the power lines along the lines of power management. Signals that ultimately will go to the mobo, where host-based firmware can ratchet overall system power consumption up or down by manipulating appropriate features on the CPU, PCH, and GPU. And even OS-level features (ie: software might slow down/pause the execution of lower-priority jobs when power budgets are tight! Such could either be power-line carrier based, or even highly accurate/high definition frequency measurement hardware (as power systems engineers know, when the frequency dips below 50Hz/60Hz, there is a shortfall of energy in a power system, and AGC in a distributed power system relies upon feedback control through frequency measurement to control generator output!).

So basically in a nutshell:

* Future PSU's will be 'smart' with respect to the power grid they're attached to.
* Future PSU's will be able to talk digitally to manage noise, temperature, etc. with the hosts, with users deciding whether they want more fan noise or if they'd prefer component de-rating.
* Future PSU's will be "configuration-aware" so they don't keep more hardware "online" than is truly necessary for the peak loads that a system might actually experience (ie: a 1000W supply might have 3-4 "trains" (some might use the term 'phase', but its really not an accurate use of the term in electrical engineering!), and if plugged into a typical consumer PC that peaks at 100W, would shut down all but 1 train). A "mini supply" might even be incorporated to supply standby loads at extremely high efficiency levels, while the big supply is only turned on through a relay if the machine is turned on. Microprocessor control and coordination with the host will allow PSU's to adjust their electronics to the current and near-time predicted future operating state of the machine for maximum efficiency.
* Future PSU's will be able to monitor and report their health to the host for maintenance and trending purposes through firmware-based self-tests. Kind of like "S.M.A.R.T." for hard drives and SSDs. If you buy a used PSU, you'll be able to find out how many hours the previous owner used it for, and how many kw-hours it converted cumulatively. Hosting/datacenters will be able to extract and use this data for billing and planning purposes!

Why is all this going to happen? Because power is almost as expensive now as actual hardware over its service life. "Consumers" largely don't even use ATX anymore -- they're on tablets, laptops, NUCs. So the ATX people, to some extent, can run a bit wild with adding very powerful (no pun intended) features to the ATX standard to keep it modern.

 

[Pink Jazz]

-12V has been gone for ages. A supply I bought in 2005 didn't have it. I'm pretty sure PCI didn't need it either. The only real "use" for it was RS-232 serial ports, and those have have had charge-pump-based MAX232's or similar available for the past 20 years. BJT TTL logic required bipolar supplies, but computers have been nearly all unipolar CMOS for at least 20 years now.

Pin B1 on a PCI slot uses the -12V rail, however, it isn't used by most PCI cards. The only PCI cards that generally seem to use it are a few sound cards where the -12V is used as a reference voltage.

 

[Red Squirrel]

A 48v standard could be interesting, for super high draw items like video cards. Instead of having all those 12v runs going to it, you could get away with maybe 1 or 2 48v runs. Could maybe use 12 awg wire as well. That would be good for 20 amps, and at 48v that would give 960w, which should be enough for most video cards and other high power cards.

I'd love to see them get 48v IN as a standard as well, make it a standard that they either accept 120/240v AC or 44-56vdc (or similar range). For better compatibility make the negative input floating, so that -48v or 48v can work. Of course there would need to be some kind of check to ensure both DC and AC can't be plugged at same time.

This standard could extend to UPSes, instead of UPSes being a standby device, they would basically be a high output 54v PSU that happens to have a battery inline. (54 is float for 48v worth of batteries)

 

[StrangerGuy]

I dunno, but by this point in time it's already way too late for a new desktop PSU standard to be commercially viable at least for the consumer side.

 

[Pink Jazz]

I actually wonder, what actually still uses the 3.3v line today? I believe that it was added to the original ATX standard primarily for Slot 1, AGP, and SDR SDRAM. Yes, I know SATA disk drive power connectors have an orange 3.3v wire, but most SATA disk drives do not actually use it as far as I know, otherwise Molex to SATA adapters won't work.

 

[Red Squirrel]

I'm guessing 3.3v is used more by the motherboard and similar components. Microcontrollers, FPGAs, the actual CPU etc. Though a lot of that stuff can run at 5v too.

I do think they could simplify it though, get rid of the negative rails, as a start. Creating negative voltage is actually more complex than simply stepping down voltage. Not even sure what uses the negative rails. Sound cards maybe?

 

[Elixer]

I'm guessing 3.3v is used more by the motherboard and similar components. Microcontrollers, FPGAs, the actual CPU etc. Though a lot of that stuff can run at 5v too.

I do think they could simplify it though, get rid of the negative rails, as a start. Creating negative voltage is actually more complex than simply stepping down voltage. Not even sure what uses the negative rails. Sound cards maybe?

Well, RS232 uses -3V to -12V for logic 1 & +3 to +12V for a logic 0, and there is still a ton of industrial equipment that uses those ports.
Also some Op-amps and DAC's uses -12V.

I rather they have a new form factor for cases & new PSU specs, and drop all the legacy stuff.

 

[Pink Jazz]

Well, RS232 uses -3V to -12V for logic 1 & +3 to +12V for a logic 0, and there is still a ton of industrial equipment that uses those ports.
Also some Op-amps and DAC's uses -12V.

I rather they have a new form factor for cases & new PSU specs, and drop all the legacy stuff.

Yes, and there is also a -12V pin present on conventional PCI (not PCI-E) slots, and was required by some sound cards (high end models in particular) as a reference voltage.

I agree that -12V will be the next to go once conventional PCI is gone after a while, and perhaps we could see 3.3V go away as well in the future, leaving just 5V, 12V, and perhaps a future higher voltage line for greater efficiency.

 

[tynopik]

i wish there was a ups spec so you could just plug a battery straight into the power supply instead of having to go through a wasteful inverter

 

[piasabird]

I would go the other way also. Maybe we need a new ATX Sub-compact power supply like SFX but smaller. When it comes to building a computer without a video card using the integrated video, using a large ATX power supply is just wasted space. The power supply is almost as large as the motherboard.

 

[dave_the_nerd]

I would go the other way also. Maybe we need a new ATX Sub-compact power supply like SFX but smaller. When it comes to building a computer without a video card using the integrated video, using a large ATX power supply is just wasted space. The power supply is almost as large as the motherboard.

PicoPSU.

One of these would be sufficient to power an Atom or ULV rig with no GPU and a single SSD.

https://www.amazon.com/Replacement-Adapter-Western-Digital-World/dp/B005PSO33W/

 

[Red Squirrel]

i wish there was a ups spec so you could just plug a battery straight into the power supply instead of having to go through a wasteful inverter

I'd love to see telecom 48v standard become more within reach. It's a neat concept, you have a bunch of rectifiers (basically AC DC supplies) that keep a battery bank on float, then all the equipment connects to that same bus. It's fully isolated from outages and other noise that can happen on the commercial AC side. You could have PSUs that have both an AC in and DC in, and the DC in would take a wide range for all sorts of setups, like say, between 10 volts and 60 volts. Not sure how easy it would be to design that though.

 

[BTRY B 529th FA BN]

delete

 

[etherealfocus]

Whence all the talk of super-high power delivery for PCs? A high end graphics card uses under 200W these days, 2 GPUs max seems to be the way of the future, and SSDs and RAM are both reducing requirements for the small stuff as well. M.2 draws power through the mobo which eliminates some cabling requirements... simplification and efficiency seems to be the way forward. A modern gaming rig this generation uses maybe what, 600W absolute tops? 2x180W for a pair of 1080s (let's call it 2x200W for some OC headroom), 150W for an overclocked HEDT CPU, and maybe 50W for mobo, RAM, storage, and accessories.

Given the absolutely tiny marketshare of rigs like that, it doesn't seem worth adding a higher voltage line just for those few edge cases... how many people dropping >$2K on a gaming rig care about saving a few bucks a year on their power bill?

IMHO the way to go is reducing PSU volume and complexity - make SFX or similar the standard size and drop all but +5 and +12 rails. Micro PCs command way more market interest than uber high end systems.

I'm not an EE, but that couldn't help but have some knock-on benefits for reliability...?

Also, yes, digital PSU status reporting a la SMART would be sweet. Especially if it goes beyond simple kWH-converted to realtime operating efficiency, component health, etc.

Is there any reliable way to simply block deficient by design cheapo PSUs from being sold under the ATX3.0 label? Would be amazing to be able to trust that anything with the label would be of a reasonable minimum quality.

 

[stormkroe]

It seems like 48v would be useless for a cpu, as that high of a voltage would be really hard to keep from leaking in transistors with 14nm gates.
Also, as I saw first hand when working for Fujitsu, the 48VDC spec in telco applications is positive ground. It's hilarious how ingrained a negative ground is in our thinking, and workers invariably get it wrong sometimes. Very sparky