A weak or dying PSU impersonates everything. GPU crashes, USB dropouts, “bad motherboard,” even RAM errors. People replace those parts in sequence and the shutdowns continue. Power is upstream of all of it. I check that before I throw a board at it.
The pattern
- Crashes when the GPU loads (games, render) but idle is stable.
- The machine needs two presses of the power button, or it comes back on by itself after a shutdown.
- A burning smell at the rear grille, or a coil whine that recently got mean.
- You added a second drive or a new GPU and the symptoms started the same week.
Paper wattage isn’t the test. A 700W no-name and a 650W known unit are not the same 12V rail. If the GPU was on a SATA-to-PCIe adapter, treat the PSU and the adapter as suspects before the card.
On the bench
A known-brand unit with honest 12V, sized for the GPU
Single mid-range GPU, I like 650W from a brand that publishes full-load numbers. Don’t buy the biggest wattage on the box. Buy the 12V rating and the connectors you actually need.
- ATX 3.0 / 12VHPWR only if your GPU needs it. Otherwise a quality 8-pin unit is correct.
- Keep the modular cables that came with it. Do not mix.
Swap as a test
Spare known-good PSU? A test swap is worth more than a software log. If you don’t have one, still rule out the wall (another outlet, no dying surge strip) before you order. Then replace once. “Maybe this 500W is fine” is how you buy twice.
Verdict
Load crashes plus a cheap or aging unit: replace the PSU before the motherboard. Size it to the GPU, skip adapters, keep the cables that came in the box.