What Size Power Supply Do I Actually Need?

Two components decide it

A CPU and a graphics card account for most of a build's draw, and both publish a figure on the manufacturer's spec page. Everything else together — motherboard, memory, drives, fans, pumps, peripherals — is usually around 100 W.

That last number isn't worth agonising over. It's small next to a graphics card and smaller still next to the headroom you're about to add.

A 125 W CPU and a 250 W card comes to 475 W peak. The graphics card alone is 53% of it.

Why the sum is the wrong answer

TDP is not peak draw. A card rated at 250 W can pull far more than that for a few milliseconds during a load change. Transients well above the rated figure are normal behaviour, not a fault.

A supply that's adequate on paper can trip its own over-current protection on those spikes and shut the machine down — and that presents as a random reboot under load, not as anything obviously power related. Which makes it genuinely miserable to diagnose.

This is what the ATX 3.0 specification was written to address, and it's the strongest argument against sizing tight. If that 250 W card briefly doubles, you're asking for 725 W from a system whose sum is 475.

The second reason: efficiency peaks at half load

80 PLUS certification tests at 20%, 50% and 100% of rated load. The 50% point exists in the standard because that's where power supplies perform best.

An 80 PLUS Gold unit is specified at:

Load	Efficiency
20%	87%
50%	90%
100%	87%

So the goal isn't just enough watts. It's landing your normal load somewhere near the middle of the unit's range.

But your normal load isn't your peak

This is the part that's usually skipped.

A system peaking at 475 W doesn't sit there. Gaming is roughly 70% of peak for most builds; desktop work is far below that.

So sizing until peak equals half the supply means buying about twice what you need — and then running at 25% typical, which is back down the curve on the wrong side.

Centre the typical figure, not the maximum.

The two arguments agree, which is the reassuring part

Approach	Answer
30% headroom over peak	618 W
Centre typical load at 50%	665 W

Both round to a 650 W unit. On that supply your typical load sits at 51% — essentially exactly the efficiency peak — and peak load at 73%, comfortably clear of the ceiling.

When a rule of thumb and a physical argument about efficiency land in the same place, the answer is probably right. If they diverge sharply on your build, question the typical-load estimate rather than picking whichever number you prefer.

What happens without headroom

Set headroom to zero and the tool suggests 500 W. Peak load then sits at 95% of the supply — running flat out, at the least efficient end of the curve, with nothing left for a transient.

That's the configuration that reboots under load.

Wattage is not quality

A cheap 850 W unit is a worse purchase than a good 650 W one for a build that needs 600.

The rating describes capacity — not how cleanly power is delivered, how stable it stays, or what happens when something fails. And the 80 PLUS badge is narrower than it looks: it certifies efficiency, not build quality or protection circuitry.

Reviews that actually load-test a unit are worth more than either number.

Two things this doesn't check

Connectors. A card may need PCIe connectors your supply doesn't have, or a specific 12VHPWR cable. This is the most common way a correctly-sized supply turns out to be the wrong one.

Rails. A multi-rail design can be unable to deliver enough current on one rail even when total wattage is fine.

Check the connector list against your graphics card before ordering — and the physical length against your case while you're at it.

---

Sources: 80 PLUS efficiency tiers and test points as published by Seasonic and Cooler Master.