When will we start seeing Power Supplies in PCI slot form factor?

I was looking at some compact Micro ATX case designs, and something that is always "in the way" is the power supply since it tends to be closer to a volumetric square compared to other components being more of a rectangle (low height compared to width and length). I was wondering when or if we will ever see power supplies in the shape of today's graphics cards. Maybe a future revision of the PCI slot would allow us to power the motherboard and CPU directly through the PCI-e lane itself, and add an additional slot in front or behind the PCI slots that can connect PCIe cards with the PCI power supply. The only cables you would need is for fans and mechanical hard drives. Probably the first PCI power supplies would be 3 slot variants until we can get the efficiency up enough for two slot versions. Hell we might even be able to put waterblocks on our PCI form factor power supplies! Seasonic/Silverstone you listening?

Just wondering if there is anything similar to what I was thinking.

:stuck_out_tongue:

1 Like

Is there 600-800 Watt variants?

Haha :smiley: no

No way to cram this much current onto this connector.

Doesn't have a whole lot of peripherals anyway. It's just to show that PSUs can be pretty small already, it's just a matter of efficiency, rated power and usecase

They are already almost as small as they can reasonably be. Take a look at how tight the guts of the Corsair SF600 are, perhaps the smallest 600 watt power supply on the market:

  • Big coils on top left and top middle
  • Power connector with line filtering caps on top right
  • Massive cap on middle left
  • Heatsink for the MOSFETs running top to bottom
  • Transformers with the Corsair labels
  • Daughterboard for the minor rails on lower right
  • Connector board across the bottom

For a bad power supply there needs to be at least a transformer, bridge rectifier, and voltage regulators to even be functional as a power supply. For a good power supply there needs to be all of that plus filtering caps, components for the switchers, MOSFETs, coils, chokes, diodes, et cetera. Some of the components need to be pretty beefcake too since they are dealing with a few hundred volts and/or a bucketload of amps.

It would be nifty to see smaller power supplies, but right now it's not possible because they pretty much physically can't make a power supply that small. Or at least one that small that won't blow up the computer.

5 Likes

Modern PSUs have come a long way in the last 10 years but they are still pretty darn big. I suspect Physics in this case.

2 Likes

its basically already been said but safety, stability, performance, output, and use-case all come in to play here...i think that the best we can hope for in the short-term is reduced heart output and increased efficiency (which of course go hand-in-hand) and possibly high-wattage units that are marginally smaller and could possibly get away without using fans...it seems to me that something you're thinking of, 600-800W in a form factor around pico-size just isnt going to happen in anytime in the NEAR future

who knows though...wish for a revolution but i wouldnt expect more than baby-steps

@avattz & @tiramisu

When dealing with high current power systems, there is no way to get around physics.
Due to the physical phenomena that coils and transformers use to work there is a physical limit to how small they can be made and still function the way they should. See induction, resonance etc. A read through the wiki page about transformers and inductors will show you that properties such as size and surface area and wire diameters are very important characteristics.
Current carrying capacity of any given piece of wire is also dependent on that wire's thickness (surface area actually), so for a given high current PSU we are already limited by the wire thickness and transformer size required to handle a given load and not catch fire.

Electrical Power systems design is a pretty complex topic and the technologies it requires do not scale as well as transistors and other low current switching electronics that are only concerned with signals.

1 Like

The best you can hope for is an adoption of rack servers' PSUs in tower cases. They're usually much thinner but also much longer than your usual power supplies. And they're either loud as hell (because of smaller high-rpm fans) or passive (when relying on in-chassis fan assemblies).

2 Likes

Probably, with the use of new materials that can work with the same properties os the ones that are currently used. Power supplies have to be that big because the need to make 110V to 240V into 12V, 5V and 3.3V and convert the 50Hz to 60Hz oscillation of the tension into a continuous tension.
First of all you need a coil with noumerous windings and one with less windings to go from a higher tension to a lower one. After that you need to have a straight tension and, to do that, you need a big capacitor and diodes to make it as straight as possible. This is just one stage, and you need two more stages to have all the tensions you need to make your PC work. These are the bare basics of an AC to DC power supply.
Maybe ii could be possible to have a power supply that makes 110V to 240V into 12V and on the motherboard are integrated the other stages.

I guess "someday" it might be possible, but given the state of PC components and the amperage draw of those components not many single connectors and certainly not a PCIe buss slot could handle the amp draw and live for very long, there is a very good reason that a Raspberry Pi has it's power transformer/rectifier at the wall outlet and a PC has a big box in the case that has some sort of cooling solution....when you move pixies (voltage) around, rectify it (convert from AC to DC) then provide a lot of amperage/wattage for devices to draw there is a lot of heat generated that has to be removed somehow.

Modern PSUs have come a long way in providing good clean (no interference / noise / RFI) consistent power that can live a good long life, a good quality PSU can last through several builds providing power in spec for years and years, a cheap PSU will die a untimely death and probably create collateral damage when it prematurely fails.

To have a PCIe slot PSU a lot of the components would have to be moved onto the motherboard creating even more heat on a PCB that has to endure a ton of heat already, one of the reasons we can have small compact computers is that components are moved off the MB either eliminated totally, consolidated in some other component (think northbridge/southbridge), or moved to add-in cards, features require space to implement (think M.2) and everything in a PC needs a connection to good, clean, power, we can get very small form factor PCs just look at the Intel NUC but it's always at the cost of expand-ability / flexibility.

They could design new power supplies with a thinner and profile. I think one of the big problems is trying to implement a new standard. It takes a lot of money to invest in something like that, and there are countless instances where companies have made modular components and smaller form factor designs that never catch on for whatever reasons and fade away.

It would make sense to make a higher wattage in the same form factor as a high end video card since many smaller designs that use such cards are limited by that size and shape. There really isn't any reason why they couldn't use a dummy PCI-E card (no connections to PCI-E pins) that just slots in next to a card and has a cooling system designed to work in conjunction with the graphics card cooling system. It might be best if a PSU company worked with a GPU company to design such a standard should they feel this market segment is worth that kind of investment.

Open source hardware designs from the community could push the manufacturers in this direction. Given how much easier it is today to find people willing to make one-off designs (3D printers, water jets, lasers and more), this could be possible. Or more likely it would fragment the same way Linux does and fall into obscurity the same way other small form factor designs in the past have done as I mentioned above (What happened to STX boards? No more new boards with Kaby Lake chipsets?) I'm not saying it is impossible, I'm saying the people and politics involved are complicated.

Then there are heat/airflow issues. Without an agreed upon form factor in the first place, with numerous companies backing it, it could turn into a disaster. Having a PSU and case company work together, just for card manufacturers to not care could cause all sorts of fitting or cooling issues. This alienates customers and everybody loses.

Going external makes more sense by removing a heat source and airflow obstructions. Unfortunately many people have some strange hate towards power bricks. There isn't any reason why someone couldn't take an ATX power supply and convert it into an external unit. Perhaps running appropriately sized wires for each power rail to the case, and making connectors to attach everything from there.

My question is more about why are so many people trying to shoehorn such power hungry beasts into a small space in the first place? I like lots of small devices for their power efficiency and I love tiny houses. I have yet to see a tiny house that was too small to fit a full sized rig in there somewhere. I'll stop myself from going on a massive tangent, but I'm not a fan of harmful societal expectations and I always choose function over form, so I am heavily biased here.

Perhaps faster external bus ports, newer external graphics card designs (since they don't need to be a card shape in an external case), and USB-PD will drastically change this landscape over the next few years. I plan on wiring my tiny house for USB-PD since I have no real need for AC current inside my house.

1 Like

PCIe power supply doesn't make sense and even if it did:

Since the computing power of the past has shrunk to the size of a raspberry pi or a mini PC, Mid-Full tower builds will survive by getting more and more powerful for the same size which has already kinda happened. Today, if you want the same document editing, video viewing and web browsing experience (in terms of performance) as in the early 2000's, you really don't need a full tower build and definitely not more than 400 W of power.

So, since that part of the target audience is slowly getting filtered out, there will be a bigger % of "enthusiast" Full Tower builds - overclockers, streamers, video editors, hoarders with a bunch of HDDs, professionals in certain industries etc. These people will probably be using 600-1200W PSU's. They will certainly not want to skimp out on a PSU as the builds themselves get pretty expensive if you need more than 700W of power. It's a self-feeding loop.

What I'd like to see is a design which will hopefully make PSUs cheaper for these builds over time.
A design that externalizes, flattens, elongates and most importantly "modularizes" the PSU, with or without detaching it from the case. To me, it's kind of stupid to throw away a PSU that you got for $150+ if the only failed or starting-to-fail component is a capacitor or even worse - the fan... just 3 years after purchase. I can imagine segmenting these expensive, efficient, high power PSUs into 3-6 small modules that basically get plugged into each other or onto a mainboard with beefy molex-like connectors.

Now that would be a truly modular power supply. The smaller modules would definitely be cheaper to service or replace down the line. With a fan replacement every few years, 70% of the power supply could last for decades without full replacement.

1 Like

With the amount of circuitry it takes to make a PSU and to fit it in its 'relatively small' surface area, and the amount of wattage necessary to power all the PC components, I highly doubt there will be a compact 'PCIe' version in our lifetime

This is my first visit to this site, and the title of this thread caught my attention. Good reading, but I was shocked to find the fundamental part of improving Power Supplies was not touched on at all, namely Power Electronics Topology and efficiency. The first fact about modern power supplies is they are quite grossly inefficient, and are founded on the Buck converter and its derivatives (forward, bridge, etc), including the resonant variants like the LLC resonant converter based on half/full bridge drivers. The Buck converter was invented during the 1950’s and improvements in high frequency switching devices, has largely hidden the failings of this topology. There have been great strides made over 60y but the fact is a lot of waste heat accompanies the power conversion process, and looking at the limiting factors in the process reveals almost without exception, all modern topologies use the air-gapped inductor for energy storage and power transfer. Inserting an air-gap into what is an AC-inductor to convert it to a DC-inductor so it may pass a large DC current, is the root cause of bad efficiency, large size and poor energy transfer capability. The hope of multi-MegaHertz switching to reduce size has been proven futile, as this also brings reduced operating core flux density, increased core losses and increased switching losses, that offset much of the size reductions.

Several problems underpin inductive energy storage topologies, in that currents are hard switched causing noise generation resulting in need for filtering, and various resonant topologies that reduce switching losses have not alleviated the underlying issue of inefficient inductive energy storage. There is much more that could be said on this topic, but suffice it to say 90% efficiency is the norm here, with slightly better efficiency obtained only at the price of greater increased size. But yet new topologies are now available that will allow the Buck converter and its derivatives to be retired - new topologies have already been invented over the course of the last decade. Several new elements have emerged that offer efficiencies in the 98% to 99.5% range which reduces waste heat by over 80%.

First and foremost capacitive energy storage is by far more compact by 2 orders of magnitude, and coupled with a new PWM-resonant switching topology allows single cycle response settling times, and zero current switching, dramatically increases efficiency overall. An inductor and capacitor resonant network is still in play, but introduction of resonance scaling allows capacitance to be increased by 2 orders of magnitude, with a corresponding reduction in the size of the inductor component, and what’s more this is now achievable at 50kHz frequency instead of MHz frequencies; which save considerably on switching losses. Power supply size reduction approaches >80% in eliminating ferrite cores and large heat sinks, with a corresponding cost reduction. For example the inductor has no ferrite core (i.e. air-cored) and becomes a 5 mm length of copper wire, while a bank of ceramic chip capacitors in parallel are minuscule compared to what is used today. These technologies also offer reduced operating flux densities for high frequency isolation transformers, that operate purely as AC transformers with no need for air-gaps to pass DC current.

Here is a link to an article describing some of these new elements.

The ATX PSU operates at 98% efficiency with a single stage PFC and DC-DC converter, generating much less EMI noise than today’s PSU. The motherboard VRM operates at 99% efficiency using just a single stage for each of the CPU, GPU, and memory components, with response settling times an order of magnitude faster than the multi-stage buck converters. The GPU itself also would use a similar VRM.

There is really no reason why memory modules could not incorporate their VRM on each memory module, so small is this new switching topology. The entire VRM is now just a single chip incorporating controller and power stage.

1 Like

I might add another reference here for a single stage PFC and DC-DC converter with 98% efficiency, using the newly discovered elements of the Cuk patents.

See section on: 98% Efficient Single-Stage AC/DC Converter Topologies
http://www.power-mag.com/pdf/issuearchive/46.pdf

I tried to find a picture… but back in the XT/286 days, there was an ISA battery backup board that would essentially do a suspend to RAM type feature. Maybe someone with old issues of PC Magazine from like 1986-1990 could find the advertisement.

Yip! Back in those days transistors were clocking at 10 kHz or less making the power supplies quite large. Today with 100 kHz to 1 MHz switching speeds the sizes are significantly reduced. In 1976 I built a small inverter to run a 100W audio system, and size was immense to say the least. But the point I was making is that even 25 MHz switching speeds will not eliminate the ferrite cored inductor, and so this places a limit on Buck converter size.

Using PWM-resonant switching and resonance scaling, eliminates the ferrite cored inductor completely and removes this size constraint. With this new topology even 50 kHz switching speeds suffice to reduce the inductor to 10 nH, which requires just a single 5 mm long copper wire. Needless to say that wire may be thick and hence draw much larger currents . And at 99% efficiency no heat sink would be required and VRM size is reduced probably by as much as 90%. At 50k Hz switching losses minute. Ceramic MLCC capacitors are only 1mm x 2mm in size requiring very little sapce also.

I hope that helps to explain why ~90% volume size reduction is possible.

1 Like

Interesting, any good ELI5 links on this?

I’ve recently been reading up on DC-DC converters and VRMs and those TI ICs that are basically VRMs with chokes integrated into the IC/package (somehow I got stranded from solar arrays). This whole field is fascinating to me, there’s so much stuff I used to take for granted.

Not really no. I work with Cuk patents directly which are terse by nature. I find it amazing that power supply manufacturers have not updated their topologies. You might be interested in this on AC-DC single stage topology.

http://www.power-mag.com/pdf/feature_pdf/1310569074_Teslaco_Feature_Layout_1.pdf
Open

1 Like