A spare PC in the corner of the workshop stopped turning on. No fans, no fan-twitch, no relay click, nothing. The kind of dead that makes you assume the motherboard until you test the obvious thing first. I had a known-good PSU on the shelf, swapped it in, and the machine sprang to life. So the fault was the power supply, and the easy answer was to bin it and buy a new one for twenty quid.
I didn't, partly out of stubbornness and partly because a PSU that fails completely, rather than catching fire or browning out, often fails for one small, cheap, visible reason. So I took it to the bench instead.
the safety bit, which is not optional
Before any of the fun, a warning I am genuinely serious about. The primary side of a switch-mode PSU holds large electrolytic capacitors charged to mains-rectified voltage, well over 300V DC, and they can stay charged for a long time after the thing is unplugged. That charge will hurt you badly and is entirely capable of killing you. This is not the place for bravado.
So: unplug it. Leave it unplugged for a good while. Then, before touching anything, discharge the main primary capacitors deliberately through a resistor, not a screwdriver. A screwdriver across a charged cap makes a bang, a pit in the blade, and teaches you nothing about whether it's safe. I use a resistor of a few hundred ohms with insulated probe leads, held across the cap terminals for several seconds, then verify with a multimeter that the voltage really is down near zero. Only then do I trust it.
If that paragraph made you nervous, good. That nervousness is correct and you should keep it. If it made you think "nonsense, I'll be quick", please buy the new PSU.
what was actually wrong
With the board safely dead, the fault found me before I'd even reached for a meter. On the secondary side, in the cluster of capacitors that smooth the low-voltage rails, one had a domed top where its neighbours were flat. A classic bulged electrolytic. Years of heat had cooked the electrolyte, the internal pressure had pushed the vent up, and its capacitance had collapsed. The PSU's own protection had presumably seen the rail misbehave and refused to start, which is the supply doing exactly the right thing.
A bulged cap is the friendliest fault in electronics. It announces itself visually, it's a standardised part, and it costs pence. This one was a 1000µF 16V on a 12V rail. The replacement rules are simple: match the capacitance, match or exceed the voltage rating, and for switch-mode supplies use a low-ESR part rated for 105°C, not the cheapest 85°C general-purpose cap, because the lifetime difference in a hot PSU is enormous.
doing the swap
Desoldering an old through-hole cap is the fiddliest part if you don't do it often. The trick is heat and patience: get fresh solder onto the old joint first to help the iron's heat flow in, then use a solder sucker or braid to clear each leg. Don't lever the cap out while only one leg is free, you'll lift the pad and turn a five-minute job into a misery.
Polarity matters and getting it wrong is how electrolytics turn into small fireworks. The capacitor has a stripe marking the negative leg; the board has a marking, usually a shaded half of the silkscreen circle, for the same. Match them. I checked twice, because I have form for soldering things in confidently backwards.
New cap in, legs trimmed, joints clean and shiny rather than dull and blobby. A good solder joint looks like a little volcano, not a ball of chewing gum.
the test, at arm's length
I don't trust a repaired PSU straight into a motherboard worth more than the PSU. First test is the paperclip trick: bridge the green PS_ON wire to a black ground on the 24-pin connector, plug it in, and see if the fan spins and the rails read correctly on a meter. 12V, 5V, 3.3V all within tolerance, no smell, no noise. I let it run on a dummy load for a while and kept an eye, and an ear, on it.
It held. Rails steady, the previously-bulged section now reading clean. Back into the spare PC, and the machine has been running since without complaint.
was it worth it
Purely in money, no. An evening of my time against a twenty-pound replacement is not an economic win, and I won't pretend otherwise. But that was never really the point. The point is that "broken" turned out to mean one tired component out of hundreds, the supply had failed safe rather than dangerously, and I now know exactly what's inside the box humming under my desk. There's a quiet satisfaction in fixing the small wrong thing rather than throwing away the ninety-nine right ones, and one less PSU went in the bin. I'll take that over twenty quid.