Sizing a Power Supply Without Guessing

Size it so ordinary load lands near half the unit's rating, then check the spike behaviour of your graphics card. That is a much smaller number than any online calculator will give you, and it lands in the efficient part of the curve with room for the transients that actually cause shutdowns.

A note on this guide: this rests on how power supplies are specified and how modern cards draw current, not on units measured here. See how we test.

The problem, as it appears

You need a power supply and every source gives a different answer. A calculator says 850 W. The graphics card's own page says "750 W system recommended". Someone on a forum says they run the same card on 650 W with no trouble, and someone else says that is reckless.

All of them are describing different things, and none of them is describing your machine.

Why the calculators disappoint

Because they sum peaks that never coincide. Each component contributes its maximum figure, the total is added up as though CPU and GPU could both sit at maximum indefinitely, and then a margin goes on top of that.

Real machines do not behave that way. A game loads the GPU heavily and the CPU moderately; a compile does the reverse. The simultaneous worst case the calculator models is close to unreachable.

The practical consequence is not danger, it is waste — and mild inefficiency, because a unit spending its life at 25 percent load is running away from its best efficiency point.

What actually matters

Measure rather than model

The site already has the honest method for this: put a meter between the machine and the wall and watch what it actually draws under the load you actually apply. How much power a gaming PC actually uses covers doing it, and it was written for UPS sizing — but the reading is the same reading.

Two caveats when using it for PSU sizing rather than UPS sizing. Wall draw includes the supply's own losses, so the DC side is somewhat lower than the meter says. And a plug meter samples too slowly to see the spikes below.

Transient spikes are the failure people misdiagnose

Modern graphics cards draw brief current excursions well above their rated figure — milliseconds long, and far above average. A supply that is comfortably adequate on average can trip its own over-current protection on one of those and shut the machine off.

The symptom is a machine that reboots under load with no error, no heat problem and nothing in the logs. People replace graphics cards over this.

The mitigation is headroom and a unit whose protection is not hair-trigger, which is a build-quality question rather than a capacity one.

Efficiency peaks in the middle

Power supply efficiency is a curve, not a constant, and it typically peaks somewhere near half load. That is the argument for landing ordinary load around the middle of the unit's rating: it is where the supply wastes least, runs coolest and — on units with a fan curve — is most likely to stay quiet.

It is also why buying vastly more capacity than needed is not free. It is not dangerous, it is simply paying more to operate further from the efficient point.

Capacity is not quality

A 1000 W unit from an unknown brand is not safer than a good 650 W one. What protects the rest of the machine is the protection circuitry — over-current, over-voltage, short circuit — and the quality of the components that have to survive years of thermal cycling.

This site's standing rule applies with force here: a component that fails silently is the signal to spend more, and a power supply is the clearest example on the whole site of something whose failure takes other things with it.

The options, and who each suits

This page contains affiliate links. If you buy through them, Ugly Fixes may earn a commission at no extra cost to you. Products are selected independently.

A plug-in energy meter suits anyone sizing anything. It converts the entire question from argument to measurement, and it is the cheapest instrument in this category.

A PSU tester suits diagnosing rather than sizing, and it belongs here because "the machine reboots under load" sends people to buy a new supply when the old one may be fine.

A surge-protected strip suits everything upstream of the supply, and is a different problem from sizing — covered properly in does a gaming PC need a UPS.

Keeping the supply you have suits a unit that is a few years old, from a brand you recognise, and adequate on the measured numbers. Replacing a working good supply on a calculator's advice is the most common unnecessary purchase in this whole category.

The mistake to avoid

Buying capacity instead of quality. The number on the box is the easiest thing to compare, so it is what gets compared, and a large number from a brand with no track record buys nothing except a larger figure. Protection circuitry is what stands between a fault and your graphics card, and it does not appear on the front of the box.

The second mistake is diagnosing a reboot as insufficient capacity without measuring. Random shutdowns under load can be transient trips, a failing supply, a thermal problem or a wall socket — and the last of those is the cheapest to rule out and the one nobody checks. The full sequence is in what goes wrong with a gaming PC after year one.