The question: how do you tell a failing power supply from a failing motherboard on an old tower server that reboots randomly under load, and what do you test first if all you have is a multimeter and a spare PSU? I have a method, but I would like to compare it with yours.
My tower runs Proxmox and is about ten years old. It has started to reboot with no warning, always when several containers are busy at once, never at idle. No kernel panic on screen, and nothing useful in the logs. A clean reset with no shutdown messages usually means the power went away or a protection circuit tripped, not that software crashed.
My lean is to suspect the PSU first, because it is the cheaper part and the more common failure. The electrolytic capacitors age, and they hold up fine at idle and sag when the load climbs. The order I would work in:
- Check the physical layer first. Reseat the 24-pin and the CPU power connector, look for browned pins or bulged capacitors on the board, and clean the dust out.
- Swap in the spare PSU and rerun the same load. This is the best test I know. If the reboots stop, you are done. It costs twenty minutes.
- Only then, the multimeter. Probe the 12 V and 5 V rails on a spare Molex connector while the machine is under load. Most specs allow about 5 percent either way, but check the tolerance for your unit. A multimeter is slow and misses fast dips, so a good reading does not clear the PSU. A bad reading does condemn it.
If the spare PSU does not fix it, I move to the board: swollen capacitors near the CPU socket, memtest for a few passes, and one stick of RAM at a time.
Two things I am unsure about. Can a meter tell you anything useful about transient sag, or do you need a scope? And is a PSU that passes under load but fails at power-on a different kind of fault?
What do you test first, and what has fooled you?