Ultra compact GAN ATX power supply delivers up to 250 Watts
cnx-software.com
cnx-software.com
It's been such a great synergy. Energy efficiency has been a concern already yes, but I really view usb-c chargers as pushing the consumerization & massification & development of this technology to the superb, common, un-exotic levels it's come down to. We just got tired of doing less good, actively explored trying to use a somewhat harder to drive switching semiconductor, & have reaped great rewards. Miniaturization mandated optimization & we all win. I really think the interchangable-parts/modularity & resulting competition that usb-c begot started this ball rolling.
(Hoping USB-PD, with it's bump from 20 up to 48V, once again proves another huge win!)
Little Box Challenge went the other direction: instead of taking AC and producing DC power, it was an inverter that took DC power and produced AC output. It had to output 2kW and be smaller than 655 cubic centimeters (cc's). The winning Red Electrical Devil's team got 3x more dense than this[2]! 240 cc's! While inverting 2kW of juice. Second place was 340ccs.
I don't know whether we'd argue this DC->AC is harder/easier than an ATX supply (AC->DC) (probably not?) but this was quite an impressive competition, almost a decade ago, that resulted in probably some of the most power-dense, highest-efficiency power conversions that humanity has ever gotten up to. For compare, this 250W unit is 212cc's. 8x less power conversion, 88% the size.
The Little Box Challenge designs were published, so there's a ton of great reading. Alas, Google frakking let the damned site bitrot off the net?! Sooo, head to archive.org[3] to read the team's writeups. From the winning CET Red Electric Devils writeup:
> GaN transistors have many very interesting electrical characteristics (low Rds_on, low Qgate and Cds, ultra low Qrr); these create technological advantages over current MOSFET and IGBT devices (small size and low production costs). Unfortunately, they also have serious drawbacks due to their very fast switching characteristics: they are challenging to drive and require sensitive electromagnetic noise management. Another pitfall is the high voltage drop due to the reverse current when the GaN is turned off. The solution selected to overcome these difficulties is to control all the GaN transistors using soft switching for the entire operation range.
[1] https://news.ycombinator.com/item?id=8070215 (271 points, 8 years ago, 138 comments)
[2] https://ai.googleblog.com/2016/02/and-winner-of-1-million-li...
[3] http://web.archive.org/web/20171006105655/https://littleboxc...
I'm not sure how much power it can handle continuously in a pure passive-cooling mode.
What sort of power an electric drivetrain can handle for a burst of acceleration lasting a few seconds versus what that same drivetrain can handle as a continuous load are two very different things.
The same even applies with combustion powered cars too, if you look at the pickup truck market you'll see massive power numbers on every manufacturer's diesel pickups. If you look at their chassis-cab commercial trucks those same diesel engines are often derated by 30-50% compared to the pickups even while being equipped with larger cooling systems. That's because the commercial variants are rated to perform to spec all day every day where the pickup version just needs to put a grin on the owner's face when they floor it off the line and then occasionally tow a boat or something.
The little box challenge was significantly harder, primarily due to the ripple requirements on the DC side. This is not an issue in the same sense in the AC->DC case, because you don't have that problem there (you have other ripple considerations that might look similar, but are not (PFC and ripple on the output)). Usually you would require a lot of capacitance to solve the ripple issue for DC->AC and the winning team had a very interesting way of solving it without the use of very large capacitors.
The LBC was misshapen because they had an odd requirement: Be single-phase. A three-phase inverter (or rectifier) naturally has low DC ripple and is both more compact and more efficient than a single-phase inverter at all power levels. All of the winning solutions had extra energy storage elements which added inefficiency and bulk which a three-phase inverter would not have needed.
With that in mind, the LBC was a little silly. A three-phase machine at the same power level would have been much smaller than any of the "winning" designs.
A high-efficiency ATX power supply has additional requirements that the LBC didn't need to meet: Isolation. This implies the use of a transformer to protect the low-voltage DC side from the AC side. They typically accomplish this by rectifing AC, using a transformer-isolated unidirectionally optimized DC/DC supply to 12V, and then provide a handful of lower-power DC/DC converters to derive the other voltage levels used by ATX. A modern unisolated DC->AC machine is effectively symmetric in its power flow, and also functions as an AC->DC machine.
By using a higher frequency, the transformer can be much smaller and more efficient. If they were doing mains 60hz AC the transformer would be substantially larger.
I forget if they also step the supply side DC voltage up too…
They do, but they don't have to. The front-end rectifier boosts the voltage up to 400-450 VDC. This way, most of the components can be optimized for a single world-wide compatible machine that is capable of rectifying 230VAC.
In principle, you could squeeze out a bit of efficiency (or cost) by producing a North American market machine that was optimized for 120V and a separate machine optimized for 208/230V. In practice, supply chain simplification reduces overall per-unit costs, so just about everything is designed as a single machine for world-wide markets. Only the plugs change. For ATX, only the cord shipping in the box changes.
https://en.wikipedia.org/wiki/Light-emitting_diode#Blue_LED
I wonder what new engineering knowledge has emerged about gallium nitride that allowed it to become such a recent force in power supply electronics.
I didn't know that gallium nitride is transparent.
It's been known for some time that GaN has a superior electron mobility number (which, effectively means you can get a better/smaller/lower impedance/faster switching device depending on which tradeoff you want). Problem is, getting reasonably good process control has been challenging. It appears some folks in the industry have been overcoming that challenge, therefore we're seeing the results of that.
Of course, they all purport to offer a warranty, but if you try to claim, they will not honor it. They will claim it's because you bought from Amazon and they don't warrant Amazon purchases (despite this being their main retail outlet, and the fact that they advertised the warranty IN the Amazon sales materials). Or something similar.
I constantly expect failures. This seems like an absurd task I'm putting them up to. But they are all still alive & doing the job.
I've had a kill-a-watt so I can see their power consumption. A large amount of their life is spent ~60 watts, but there's definitely hours on end & heavy usage days, where they go up to or stay at 90+.
My anecdotal data looks different than yours.
Yes, there is an EU law that says retailers are responsible for 2 years(6 in a some cases) for any item they sold......but only to rectify any manufacturing defects. The law additionally states that any defect found in the first 6 months is presumed to be a manufacturing issue, anything after the first 6 months it's the responsibility of the customer to prove that the product has failed due to a manufacturing problem. I'm sure you can appreciate that this can be very hard to do.
It's not a "warranty" in the sense that most people mean warranty. If your laptop fails 18 months after purchase, the retailer only has to fix it if you can prove that it failed due to a manufacturing defect. It's different to manufactuer's warranty, which usually covers any defect no matter the reason.
And yes, there is also the law that states that products should last a "reasonable" amount of time - but what is reasonable is not a strict definition, and again, unfortunately sometimes has to be argued in small claims court. It's not so straigtforward.
I have used warranty frequently after 6 months and never had to prove it is a manufacturing defect. Of course, if you buy at some shady webstore or known-bad retailer, you might have more issues.
Some of the power-tool brands are beginning to make USB-C devices, those chargers should be reliable also. Here's Dewalt: https://www.dewalt.com/product/dcb094k/usb-charging-kit
Companies selling under brands want a premium for them (Anker, Aukey) and "no name brands" you can actually put a little confidence into (Ugreen, Blitzwolf) are, for some reason, rare.
So, you charge your laptop at 100W means 20V*5A, then connect a phone that can only negotiate to 9V. Now your laptop can, at best, only get 45W. But laptop and phone might also have to share the same 5A, meaning something close to what you described. (or, your laptop might not support 9V ... in which case I'd guess one of the devices would disconnect?)
You might be better off using a USB-A port if the charger has one, then your laptop could perhaps remain at ~100W and your phone might still get ~10W. (Assuming charger and phone doesn't also have quickcharge which might complicate it further). Or, perhaps charge your phone via your laptop (yes, negates the point of a multiple-port charger, but if you are in hurry - otherwise, slow charging is preferably anyway)
A good review will test this, and a good product page will specify this (but if your are not looking/thinking about it it is easy to miss).
I also have a 60W Satechi travel charger that's been powered-on 24/7 for 3 years now without the slightest hint of a problem. It's even been buried under cushions for hours at a time.
The only charger I've had actually die was a Hama-branded device. Hama is the cheapest possible Chinese junk the brand can find.
They’ve been selling a 400W (still fanless!) DC-DC version for years and it’s only 160x51.5x26mm : https://hdplex.com/hdplex-400w-hi-fi-dc-atx-power-supply-16v...
I use it to power my CPU+mobo but I also have a couple of server PSUs further away to power my 3090 over several pairs of THNN. It’s far from a SFF build, but it can be totally silent.
This in no way meets the atx form factor for power supplies, neither the original, nor the 12V version. Like I could forgive it if it failed to be a rectangular prism in favor of having more open space, as long as it had a frame with proper mounting points.
And I could also forgive it if they went with the rectangle, but far shorter than the official 140mm depth (that has always been mandatory per intel specs), given that the market already accommodates non-conforming lengths, and this could be made to not require the bottom of case mounting holes, allowing for absurdly short lengths.
But the photos show none of that, and just a fully custom form factor, without any apparent mounting mechanism.
This is an ATX power supply. It is just designed for very small cases that would never be able to accommodate the existing standard configurations.
Did you just pull this out of nowhere?
Look at ATX Specification v. 2.01 section 4.1, page 18-20. It specifies the "required overall dimensions and general form factor of an ATX power supply," with or without special airflow ducting (emphasis in original).
The linked PSU in the OP clearly does not meet this form factor.
Perhaps they removed it in a revision? That seems like a silly idea…
Can someone explain to me why 48V? Is there something special with 48V or beyond? The reason I ask is because some cars now have electric turbos and the manufacturer had to switch to a 48V system. Just curious what the benefits of higher voltage is and if 48 is somehow special.
Thanks in advance.
I seem to recall that somewhere around 50 volts is a magic number from a safety/regulatory perspective. It might also have something to do with semiconductor breakdown voltage, but maybe I'm out of date on that.
Specifically, 50 volts is approximately the threshold above which electricity can pass through the skin and shock you. Lower voltages will only shock humans if they have a way to bypass some of the skin, like wounds or tight-fitting metal jewelry. Otherwise, the resistance of skin is high enough to prevent the flow of electricity. That's why you can touch both terminals of a 12V car battery at the same time and almost certainly not be shocked.
One of the things that can “get you” with these higher voltage systems is you can’t use them to jumpstart the present majority of cars. While I’ve yet to actually see a higher voltage car, I would hope they’d put warnings by the battery to remind the user that “hey, don’t try to jumpstart your friends car with this battery!”
s/GAN/GaN/
[1]: https://en.wikipedia.org/wiki/Generative_adversarial_network
(the original script was written with the sensible "using humans as a biological computer" angle but they worried audiences wouldn't get that newfangled computer thing in the basically prehistoric year of 1999, so they changed it to be humans generating power which... isn't how thermodynamics works!)
"And where did you learn about physics, Neo?"
"..... In the Matrix...."
been kind of a crappy couple weeks for me due to family illnesses and it did bring a smile to my face. feels like a "the matrix has you" sketch from the flash player days...
> Neo: "You won't do it, you need humans to survive."
> The Architect (phrased angrily and bitterly) "There are some levels of survival we are prepared to accept."
I wonder if it is possible to run more than one board off the same PSU. I know you can run a board off multiple PSUs, but is the inverse true? If i ever get a new batch of these 25W TDP computers racked, having 2 of these connected to two UPS and 6 computers might be cool. It's just the ATX connector that needs population, i use NVME whenever possible, no GPU, no cards, just USB and ethernet onboard.
Absolutely. There's something I can't remember about signal lines that need to be set up, but yet, people have been running multiple computers off as single supply for a long long long time now.
Also worth warning people ahead of time, it's hard to predict what kind of specific power consumption a motherboard will have. There's so many power rails going in. 3.3V, 5V, 12V, -3.3V... Making sure that each rail is adequately rated for additional motherboards can be a challenge. Ideally one would test each supply line, one by one, to try & figure things out, & compare against rated specs of the power supply.
Intel's been pushing a new spec called ATX12VO, short for 12-volts-only, which does away with all the other rails & has the motherboard expect 12v. This would make it much easier to understand what power consumption is. I do seem to recall some of the new ATX specs (3.0?) also kind of expect power-consumers to declare how much power they expect to use, as a safety/planning feature, which could complicate this one easy-to-do hack (unless there's some way for the additional motherboard/motherboards to also declare their power draw). This might only be ATX3.0's new 12VHPWR connector, which works alongside PCIe 5.0-- not sure if it also impacts motherboards or not.
Buy a cheap inline tester and leave it always in? Something like: https://www.google.co.nz/search?q=Power+Supply+Tester+Digita...
So basically I wouldn't worry about motherboard compatibility due to 5v, 3.3v, and -3.3v rails - there just isn't much of anything still being done on those rails, and the things that do use it pretty much are the same across motherboards (because they're things that are plugged into the motherboard, not the motherboard itself).
The most important 5V rail is 5VSB - it powers wake-on-LAN and on some systems wake-via-bluetooth. The interesting question is, is it more efficient to always let 12V->5V regulators run for 5VSB or to use an extremely small separate regulator?
My main appreciation for ATX12VO (12 volts only) is the MB can choose to right size the 5V and 3.3V rails appropriately. M.2 slots are driven from 3.3V only (even if M.2 devices step it up or down for their specific uses). A board with 4 M.2 slots will have a different potential 3.3V power demand than a board with only 2 slots. Typically, a decent board will have 1 M.2 2230 A or E slot for WiFi, and 1-2 M.2 2280/22110 M (and rarely B) slots for M.2 storage. PCIe x16 lanes will have 9W of 3.3V, as part of their 75W total power (the remaining 66W is 12V).
As noted earlier, 5V is heavily I/O driven. A board with 15 USB ports can expect to need more 5V than a stingy board with only 8 (both are accounting for front I/O headers).
I'd rather the MB have to right-size the 3.3V and 5V rails, than forcing every PSU to include copious amounts of 3.3 and 5 just to cover every edge case.
Sure, it's a tiny bit of potential efficiency. It's also passing a bit of cost from the PSU to the board itself. Sadly, it does seem ATX12VO's main impact will be lost. OEMs are going to continue with their custom PSUs with unique connectors and the DIY vendor + consumer market has largely turned their backs to this.
To complicate this further, full system power-off is not the only time you want to do wake-on-lan... if you want to wake from system sleep then you have to be running the 12V rail anyway, because that 12V rail drives the VRMs which drive Vcore and Vmem. No 12V rail means you lose your memory and CPU state...
As such I don't think you can really frame this as "having a separate 5V regulator with the 12v rail turned off", or at least that use-case must exist in addition to the deep-sleep use-case, it cannot supplant it entirely.
Anyway though, power-supply 5V rails are usually much much larger than they need to be, and that has higher loss inside a transformer. Even if you make that a solid-state thing inside the transformer, moving power at 5V is less efficient than moving it at 12V and stepping it down at the point-of-consumption. So actually the answer is probably that it's more efficient to always run a step-down regulator.
When you remove mains power it switches over instantly enough that the computer doesn't notice. There's no serial port on it so the computer has no way of knowing that it's on battery - but it was also only $26 shipped. No wires or battery or anything. I happen to have a few of the trimpot breakouts specifically for setting voltages, so i just made sure that when the PC was powered it was getting the correct voltage without sagging below 12.2ish under load at the output terminal of the PSU. It also has a little trimpot on the board for adjusting the voltage.
P.S. I'm not sure if Drok has a lot of knockoffs or not, i do know that i like that brand well enough, never had anything bought from them break in a weird way. I have slagged a few of their "waterproof" 24VDC boost converters, though!
Generally the most You can get from a PSU this size would be those forementioned USB-C PD bricks. Those would these days get You at max about 100W for probably not less than 50-60USD (cheap Chinese models, more if branded).
Multiply that 2.5x (250W instead of 100W) and suddenly You're pretty close to the price.
It's a little baffling why they aren't more widely used, they aren't that complicated. https://u.dianyuan.com/bbs/u/17/1086016427.pdf is an example. It's almost comically simple: connect a current sensor, a high-pass filter, and an opamp controlling a transistor. With some thought it could probably be cut down even more.
I think TI has integrated this into their controllers already.
Sorry, pet peeve of mine.
The caps you need for the active filter are the tiny solid ceramic ones that are basically bulletproof except in high vibration environments. The ICs needed are similar.
Also < 0.4W idle. Wow.
Then again, 20kWh here, 20kWh there, it adds up.
Edit: Yes, the "ATX12VO" spec.
I have a desktop HP that has a standard-sized PSU but has a weird connector that only plugs into the motherboard. The drives are powered from a separate connector that also plugs into the motherboard. And the PSU - MB connector is much smaller than standard ATX.
There are also boards like this that have the RAM mounted flat, a lower profile set of ports on the back, and take in 19v DC: https://www.asrockind.com/en-gb/IMB-193
You'd have to make something custom, but I think it would be much simpler than typical heat sinks. With air cooling, the vertical placement would be ideal, since convection will tend to suck cool air from the bottom and release it on the top. (I think the iMac does/did this). Even for water cooling, it would be simpler with far fewer turns in the pipe. Plus, I just like the idea of laying everything out flat instead of the weird 3-D origami that is modern PC design.
A lot of prebuilt OEMs and consoles and not to mention laptops already use this approach.
On the other, my issue is that motherboards are already one of the least reliable components in your PC, and introducing even more complications to them just makes the situation (and the cost of replacement) worse
No that's silly, we should continue to use a design from 1995, itself replacing a design from 1981.
You can buy a nice AMD Ryzen motherboard for like $80 retail (Newegg shows $59 is where new unopened ones start). Normal electronics supply chain and shipping margins means it cost roughly half that to manufacture it. So on a $40 bill of materials and labor cost if you add even just $1 to make something better/modular/whatever then you're adding a significant amount in terms of percentage cost increase.
As much as reducing electronics waste is a worthy goal, the economics of it aren't aligned quite yet in today's world.
I can see this being really good for (for instance) boxes like my fileserver which has a 65W TDP i3 in it and a bunch of SSDs, or compact media boxes that just have a processor and iGPU.
Unfortunately the way things are going in the world of PCs, you aren't going to power much of a modern dGPU from that.
As a quick rule of thumb for huge transistor count ICs: Power (heat) = freq*Voltage^2
I won't go into detail, but you can see how for a CPU designer decreasing voltage is critical to increasing speed. Remember a decade ago when CPUs were like 1.2V and we hit this ~3GHz "wall"? But now due to breakthroughs in process nodes, CPUs are running I think ~0.8V but boosting up to 5GHz.
If we make the same CPU on a GaN process, the transistors would probably need ~3V to switch, and thus need to be run about 9x slower to meet the same package power (or I guess you could run it at the same speed but then try and figure out how to cool a 9x higher TDP CPU).
Furthermore, GaN is difficult to manufacture into a monocrystaline perfect lattice compared to Silicon. You can image that for modern 7nm/5nm nodes, where there are sometimes angstroms of distance in a transistor, having a foundation full of defects can be devastating to the yield.
Power electronics is a totally different domain with different challenges. In this area GaN excels, but for ICs I don't think it makes sense. Let's see in 20 years though. GaN is relatively new compared to Silicon and might just need some further refinement.
Just my 2c though. Someone correct me if I'm wrong.
Of course GaN is a much newer technology too so I imagine it'd have some serious catching up to do before it gets to the scales where it'd be useful for CPUs.
There are two factors:
- you need a high breakdown voltage, so that for a given voltage you want to switch you can make the transistor channel as short as possible; shorten the physical distance that charge carriers have to flow in the semiconductor.
- you need a high maximum drift velocity, which governs the rate at which charge carriers can flow through the semiconductor
Those contribute to reducing the on-resistance of the transistor, thereby decreasing its heat output for a given size.
There are a lot of case options for small Mini ITX systems, I'm using an NCase M1 which is a 12-13L by volume case. There are smaller cases, but the smaller the case, the less flexible it is with configurations, which is one reason they large ATX cases are still popular; it seems like a lot of enthusiasts like clear side panels and rgb light shows and so forth, and a tiny case isn't as exciting.
There are also multiple PSU size options. The NCase M1 I mentioned can use a regular ATX PS, but an SFX power supply fits better; SFX seems like maybe half the volume of a regular ATX PS. There is also SFX-L, which is a bit bigger so it can house a larger fan. These smaller PS standards are typically lower power, although you can get 650W+ SFX supplies now I think. Silverstone and Corsair were regarded as having the best SFX supplies last time I was looking.
Finally, there are a couple of options for external power bricks like you suggest. As mentioned elsewhere there is a PicoPSU which is a DC/DC converter the size of a 24 pin ATX socket. It just plugs into the ATX socket on any regular ATX-compatible motherboard and has some cables attached to the sides for peripherals. There's also the 'Thin Mini-ITX' standard, which is I think intended for digital signage and industrial stuff, but if your requirements are modest you can use these for a desktop also. They integrate a DC/DC converter and use a 19V laptop ac adapter. The ones I've seen are best suited to on-chip graphics and m.2/msata drives, though.
It doesn't help much that you can have a small motherboard. You will need a CPU cooler stick out the middle of it, which already takes most of the space you have above MB's footprint. If you insert a GPU without a riser cable, you'll have to move the PSU to the side of the MB, and that's already almost as chonky as a regular ATX. So you're forced to deal with airflow-choked GPU, low-profile coolers, and no room for a powerful PSU.
If you instead opt for an easy-to-cool CPU and no discrete GPU to save space, it's still a massive box for a computing power of a tablet.
I remembered earlier there is an alternative GPU form factor for laptops called MXM. You can get desktop boards that accept these, I think they are called Micro-STX or Mini-STX. They are small, and the GPU is a flat module that plugs in more like a daughterboard than a PCI card. You're still looking at low-power CPU though.
I'm not sure what your vision looks like, to be honest. I think you want a laptop without the integrated screen and keyboard. That's a NUC, IMO; if an integrated or laptop-grade GPU isn't good enough, you'd want to do eGPU over thunderbolt I guess.
But I personally would much prefer having to lug around a marginally bigger box instead of two smaller ones. Fewer cables to trip over, etc.
Also, if the case is a bit bigger, it will probably be able to house a bigger cooler, so it should be quieter, if I'm looking for a powerful computer.
If I'm OK with a regular middle of the road setup, there already are plenty of options from OEMs. I'm thinking HP Desktop Minis. Dell, Lenovo and Fujitsu also have similar offerings, though I'm not familiar with them. They can be a bit noisy under load, though.
Unless... the wiring is over ~30 years old, in which case it's Black + Red for Neutral and Live.
https://en.wikipedia.org/wiki/IEC_60446#Protective_earth_and...
Other small and dense terrestrial microwave and millimeter wave things use gan power amps too.
Grid operators typically have laws in place saying how many harmonics you're allowed to inject, and for stuff under a few kilowatts it's totally fine to just have a bridge rectifier straight to a capacitor. That's what every other electronic device in your house will do.
If you're making a high efficiency power supply, you'd probably want that to be a synchronous rectifier to get 1% or so more efficiency, but it would still only be drawing current on the peaks of the sine wave.
But I don't see why you'd want to go to much effort to make your current draw a nice sine wave rather than a series of peak-chopping spikes. Is it to save money on copper in the power supply wire perhaps?
The EMC Directive mandates compliance with IEC 61000-3-2 which certainly requires you to do a little bit more than hook up a bridge rectifier to the grid.
You fix a PSU cause you know what you're doing and have the time, not because it's economical to do for someone else
Why is it that I've never heard of a power supply repair technician in a data center?
If it costs $20 to fix a $100 power supply, why would you replace it?
I don't think small cost savings like that are going to be implemented simply because it makes the operation more complex without landing any manager a big fat bonus.
Unless I'm grossly underestimating the failure rate of those supplies?
I wonder if the vendor would void the warranty and that would have any consequences.
Does you DC get around 125 failures per month?
For PSUs >1000 watts, I found that server supplies are cheaper both on a relative (more watts / dollar) and absolute cost basis than ATX form factor supplies.
There's probably more legitimate demand on the commercial side for >1kw PSUs so they can enjoy larger economies of scale.
- If your hardware hasn't reached EOL, it's probably under a warranty or support contract. So you just let your provider swap/fix it.
- If you tinker with anything and there's a problem with the rest of the hardware, they'll probably won't honor the warranty or support contract, or you could have problems with your insurance.
If it were a home user fixing a PSU... that's a different thing.
'Anyone' could be a penniless student and his time costs little, specialist may charge $40 per hour.
Also even ideal free market has cost of transaction, discoveravility, etc.
Hospitals have repair technicians, but why would you need one spesifically focusing on power supplies - he does not need different training or equipment
I’ve torn mine apart and replaced fans in them and such but I’ve never had to resolder any parts onto them. If a PSU dies - it tended to do it violently and was so cheap that it wasn’t worth doing it. All the name brand ones I’ve had never had any issues and same for the dozens of other people I knew. (Or computers I worked on or built or sold to others… or when I did IT and saw literally thousands of computers…)
Besides the capacitor only costing 10 cents and a new one around a hundred bucks the advantage for me was that I had fixed the whole thing within na hour or two including diagnosing the issue and the bike ride to the shop.
I was back on my PC that same night, and 5 years later that same PSU now well over ten years old is still going strong.
And they are standardised. Makes more sense to just replace them.
I don't necessarily need a smaller power supply, but options are nice.
I think the sweet spot of GaN is in wall adapters that deliver 18-45W over USB-A/C for mobile devices rather than PSUs where the size/weight doesn't matter as much.
This 'ultra compqct' only offers 1.18 watts/cm^3 in 2022.
While this might be the best on the market, it is far from what is technically possible.