Even if you're a "computer person" who is comfortable messing with the internals of a PC, be very careful with the power supply. Opening it up should really be left to professionals. Capacitors can remain charged for a long time after disconnecting the power cord. If despite all the warnings, you open it up to fiddle with internals because you think you understand the components and can pay attention not to touch things, at the very least, use only one hand at a time to avoid creating a circuit through your heart.<p>In most other parts of the computer the worst that can happen in most cases is that you damage <i>the computer</i>, the power supply is different, there you also have a good chance of getting shocked.
I had some contact with PSU designers at a name brand server company and learned some things:<p>1. Electrolytic capacitors: Only Japanese brands can be trusted (this sounds racist or something, but it's an actual requirement). They all should be de-rated for reliability (have 2x the voltage rating). Only exception is the main bulk filter capacitors, where your only reasonable choice is 450V, but these should be 105C. Tantalum capacitors: not allowed.<p>2. Main switch MOSFETs: only Infineon can be trusted. Single active clamp / single switch topology is a no-no due to bad experience. (I think the one in the article uses a diode clamp, so should be OK).<p>3. Standby supply needs its own fuse. The problem is that the main fuse is too big for the standby supply. If there is a standby supply fault, you will fill the machine room with smoke and your brand will be mentioned in the news when machine room is shut down. This destroys shareholder value.<p>The current limit for any exposed 12V rail is only 20A or something line that (240VA UL limit), so you must not have exposed 12V when the server cover is open.<p>There was basically a 100 page requirements document all along these lines.
Some components, mostly capacitors, have the wrong hole spacing for their pins, and leads were bent to fit. That suggests part substitutions during assembly. Many of the through-hole components were not pushed all the way down into the board. That suggests overworked assemblers. This was obviously hand-assembled.<p>The author makes a point that there's a clear division between the AC line side and the DC side. That's required for UL approval, but power supplies do show up without it. The power transformer that crosses the boundary should also have a split bobbin, so the AC line side coil and the output side coil are separated by a barrier, not wound on the same bobbin.<p>There are worse power supplies.[1]<p>[1] <a href="https://linustechtips.com/topic/515473-teardown-and-review-of-the-turbolink-atx-xg500w-crappy-psu/" rel="nofollow">https://linustechtips.com/topic/515473-teardown-and-review-o...</a>
On a somewhat related note, I just found out johnnyguru, a major contributor to consumer psu reviews and knowledge, has shut down his site and forums. His tear downs were always fascinating<p><a href="https://web.archive.org/web/20210126113732/http://www.jonnyguru.com/forums/archive/index.php/t-16610.html" rel="nofollow">https://web.archive.org/web/20210126113732/http://www.jonnyg...</a>
> The secondary circuitry produces the four output voltages: 5 volts, 12 volts, -12 volts, and 3.3 volts. [...] The power supply also provides a negative voltage output (-12 V). This voltage is mostly obsolete, but was used to power serial ports and PCI slots.<p>And even older PC power supplies also provided a -5V voltage, the corresponding pin on the ATX connector is now (according to Wikipedia) a reserved pin.<p>This progression also shows in the expansion slot standards: the ISA slot had pins for -5V and -12V; the PCI slot removed the -5V pin; and the PCIe slot finally removed the -12V pin. That is, a motherboard without any ISA or PCI slot, and without a RS232 socket or header, has no use for the -12V voltage.<p>And there's already a newer power supply standard, called ATX12VO, which simplifies the power supply by providing only 12V (and a separate standby 12V). There's already at least one motherboard built for that standard: <a href="https://www.anandtech.com/show/15763/first-atx12vo-consumer-motherboard-the-asrock-z490-phantom-gaming-4sr" rel="nofollow">https://www.anandtech.com/show/15763/first-atx12vo-consumer-...</a>
Just an observation from years building computers. Never cheap out on power supplies. Always buy one (don't use one included in a case), and buy one rated 25-30% above what you think you'll need. Some of the most aggravating troubleshooting you'll ever endure can be avoided entirely.
> The 60-Hertz AC (alternating current) from the wall oscillates 60 times a second, but the power supply needs steady DC (direct current) that flows in one direction. The full-bridge rectifier [1] below converts the AC to DC.<p>[1] <a href="https://www.youtube.com/watch?v=sI5Ftm1-jik" rel="nofollow">https://www.youtube.com/watch?v=sI5Ftm1-jik</a><p>I just knew which video is linked without clicking. ElectroBOOM is a treasure.
ATX12VO[1] is a proposed major change to PSUs that hopefully will come in the near future. It will mean simpler (cheaper) designs and improved efficiency. I hope that the DIY market adopts ATX12VO before my next upgrade. Currently only OEMs have the ability build PCs with these kinds of PSUs.<p>[1] <a href="https://www.intel.com/content/dam/www/public/us/en/documents/guides/single-rail-power-supply-platform-atx12vo-design-guide.pdf" rel="nofollow">https://www.intel.com/content/dam/www/public/us/en/documents...</a>
Bear in mind this is a very old power supply.<p>A modern supply would use the same principles, but would probably have mostly surface mount components, a single controller IC, and probably separate switching supplies for each different supplied voltage.
I'm glad to see Ken's blog on HN. He does incredible stuff with reverse-engineering old proprietary chips, e.g. the HP nanoprocessor which he reverse-engineered from masks then wrote a disassembler for.<p><a href="https://en.wikipedia.org/wiki/Hewlett-Packard_Nanoprocessor" rel="nofollow">https://en.wikipedia.org/wiki/Hewlett-Packard_Nanoprocessor</a>
<a href="http://www.righto.com/2020/09/hp-nanoprocessor-part-ii-reverse.html" rel="nofollow">http://www.righto.com/2020/09/hp-nanoprocessor-part-ii-rever...</a>
Nicely written up. It's not that complicated when you start to think about the different stages, and I think this walk-through breaks it up clearly.<p>If you continue further on to the motherboard, the CPU has a multiphase SMPS (sometimes exceeding 20 phases!) that drops the voltage further to core and IO voltages (usally ~1.8V or less for the former) but still has to generate lots of Watts (think about a Xeon or Ryzen at 3+ GHz).<p>What's really neat IMHO is how Apple changed the <5W adapter market by moving to dense SMP in such a brilliant form factor. It can take 120/240 because it is a buck converter and doesn't care about the input voltage.<p>I've wondered, if the SMPS circuit in the 5W apple adapters be scaled up to 600~800W would they be enormous and/or less efficient? Because it seems like the circuits in this board are either designed for higher power so big honkin' coils & caps are necessary, or are they that big (and inefficient) in order to save money?
I'm still boggling at the way that power is not conducted electrically across the power supply but is instead air-gapped and pass magnetically, via transformers.<p>Isn't that a lot of power?! A modern PC can easily draw 500W or more. Does that mean there's some enormous magnetic field in my power supply? Why doesn't that pose a problem for the things around it?<p>(I cheerfully admit to being very very good at manipulating bits, CPU opcodes, and TCP/IP packets and knowing next to nothing about electrical engineering.)
Server power supplies are impressive. I pulled out some PSUs from old scrapped servers. In a smaller form factor than ATX they deliver more than 1kW power. You read that right. 100A at 12V actually.
If anyone wants to see a really top quality PC power supply, for comparison to a cheap one, take a look at the board and layout on a $100 seasonic power supply. Well known Taiwanese power supply company that doesn't try to compete on cheap prices and has been around for 20+ years.
A related article was recently posted on HN:<p><a href="https://spectrum.ieee.org/computing/hardware/a-half-century-ago-better-transistors-and-switching-regulators-revolutionized-the-design-of-computer-power-supplies" rel="nofollow">https://spectrum.ieee.org/computing/hardware/a-half-century-...</a><p><a href="https://news.ycombinator.com/item?id=27257448" rel="nofollow">https://news.ycombinator.com/item?id=27257448</a>
DiodeGoneWild (danyk.cz) has a bunch of very good and in-depth videos on these:<p>How Does a Computer Power Supply Work (ATX): <a href="https://www.youtube.com/watch?v=Cur3nQjjyyo" rel="nofollow">https://www.youtube.com/watch?v=Cur3nQjjyyo</a><p>How To Repair a Computer Power Supply (or other switching power supply): <a href="https://www.youtube.com/watch?v=HcYFbCqM61g" rel="nofollow">https://www.youtube.com/watch?v=HcYFbCqM61g</a>
The real treats were the bottom-of-the-barrel PSUs. I remember having to deal with a slew of these at work back in the early aughts. I remember turning on a machine one day... <i>sssFWSSSSSSSHH</i> and the magic smoke came pouring out. It would up taking out the the motherboard and hard drive with it. I autopsied the dead power supply, and it was chock full of capacitors (some blown-out) marked "Rulycon" (note, <i>not</i> Rubycon).
As I recall, the Apple II was fairly advanced for having had a switching power supply. I understand in fact it was practically unheard of in consumer electronics.<p>I know the Apple I was a kit and came with no power supply but every one I've ever seen the hobbyist has added a clunky linear power supply with a monster transformer and giant electrolytics.<p>It's almost like the two computers stood so near but just across from one another on two sides of some threshhold.
I never realized the main transformer element worked on rectified AC. Could someone explain why we rectify the AC power before the transformer? I always thought transformers worked on AC power anyways.
I just recently made a nice little 3.3V, 5V, and 12V power supply for my electronics bench using the power supply from an old server.<p>Saved me from buying one <i>and</i> kept the power supply from ending up getting throwed out.<p>"Win-win", as my annoying ex-boss would say.
I wonder where the losses are, and if GaN power transistors for switching would help with efficiency here.<p>I like Seasonic's fanless series, but wouldn't mind getting a bit more power out of the same size PSU.
Lenovo has done something interesting in recent generations of laptop power supplies that made them a lot smaller and lighter.<p>The 170W power supply for my old ThinkPad W520 measures about 17cm x 8cm x 3cm and weighs about 580g.<p>The 170W supply for my new X1 Extreme measures about 14cm x 8cm x 2cm and weighs about 320g.<p>Does anyone know what technology let them make the power supply so much smaller and lighter?
Isn't there a way to create a rectifier with just three diodes? I remember having come up with a way to make one once, but there must have been something wrong with it (I know nothing about electronics). Now I can't remember how it was, or maybe it had 4 and just looked slightly different from the typical one.
Often PSUs will have different wattage sets but seem otherwise identical eg 550/650. Are they binned like CPUs? Or are there higher rated components
The description of the regulation of the 3V3 circuit is IMHO misleading. The inductor isn't a magnetic amplifier (multiple coils would be needed), but rather a filter element.