A gaming PC that restarts during heavy scenes is often blamed on the graphics card. The usual cause is a power supply that cannot absorb momentary spikes.
Rated power is an average, not a ceiling
The wattage printed on a graphics card describes typical sustained draw under load. Actual consumption fluctuates constantly as the workload changes between frames.
Those fluctuations include very short spikes, lasting a fraction of a millisecond, that can reach well above the rated figure.
A power supply chosen to match the average may be fine for hours and then fail during a scene that produces an unusual pattern of spikes.
Protection circuits act faster than the spike passes
Supplies include over-current protection that shuts the unit down when draw exceeds a threshold. That circuit responds quickly enough to catch a transient.
From the user's side the machine simply powers off with no error, no crash log and no warning, which is why the fault is so often misdiagnosed.
The system usually restarts normally and passes every test, because nothing is broken. It only fails when a specific load pattern recurs.
Headroom is cheaper than diagnosis
The standard remedy is a supply rated comfortably above the calculated total draw, which leaves room for transients without approaching the protection threshold.
Higher capacity units also tend to run closer to their efficiency peak at typical gaming loads, so the extra capacity is not purely wasted.
Quality matters alongside capacity, since a well-designed unit holds voltage more stably under sudden changes than a higher-rated but poorly built one.
Connectors and cables introduce their own limits
Each power connector carries a defined maximum, and splitting one cable across multiple connectors can push a single line beyond what it was designed to deliver.
Using separate cables from the supply for each connector distributes the load, which is why manufacturers specify this rather than leaving it to preference.
Adapters bundled with cards are engineered for the card in question, but reusing an old adapter with a newer card is where many connector problems begin.
Older supplies degrade quietly
Capacitors age, and a supply that delivered its full rating when new may fall short after years of use without any visible sign.
A machine that ran a previous card fine can therefore become unstable with a new one even when the arithmetic suggests the supply is adequate.
Treating the supply as a component with a service life, rather than as permanent infrastructure carried between builds, avoids a category of faults that are difficult to trace.