HOW IT WORKS

Last updated: October 11, 2026

The wattage number

Add up your GPU's rated draw and your CPU's rated draw, then add roughly 150W of overhead for the motherboard, drives, fans, and anything else pulling power. Then we add 20% on top of that and round up to the next common PSU size: 450, 550, 650, 750, 850, 1000, 1200, or 1600 watts.

The extra 20% matters for two reasons: a power supply runs most efficiently in the 40–60% load range, not maxed out, and GPUs draw brief transient spikes well above their rated wattage — headroom absorbs those without tripping the PSU's protection circuit.

Graphics card figures are the official board power of the reference card. Factory-overclocked partner cards are often rated higher. If yours is, check its spec page and size up a tier if your build lands close to a line.

Where the CPU numbers come from

The TDP printed on the box isn't the most a CPU can pull, so the calculator uses the stock power limit instead:

Motherboards that remove power limits out of the box can push Intel chips past these numbers. If yours does, size up a tier.

Dual-socket server Xeons (Xeon Scalable) aren't listed. Those boards take two CPUs and usually ship with their own server power supplies, so a single-CPU calculation would come out wrong.

The connectors

Every build needs two constants no matter what parts you pick:

24-pin motherboard4/8-pin EPS12V (CPU)

CPUs rated above 300W (the big Xeon W chips, W-3175X and i9-9990XE) get flagged for two 8-pin EPS12V cables, since their boards usually need both.

On top of that, your GPU needs its own power depending on generation:

If your PSU doesn't have the newer connector natively, cards usually ship with an adapter that combines two to four 8-pin cables into one 16-pin plug (four on an RTX 5090) — check the adapter's rated wattage matches your card before relying on it.

Want the details for a specific card?

The guides run the same math for popular cards and common CPUs, like the RTX 5070, the RTX 5080 and the RX 9070 XT.

What this tool doesn't do

It doesn't account for overclocking headroom, multiple GPUs, or exotic peripherals. Treat the output as a solid baseline, not a hard ceiling — when in doubt, size up a tier.