- Measured, not rated. — Brand calculators
add up rated TDP. Modern GPUs draw close to (sometimes above) their rating
in games and spike far past it for milliseconds — TechPowerUp's charts
show a 300 W-rated RX 6800 XT drawing 295 W in games but spiking to 567 W
for 20 ms, and an RTX 4090 rated 450 W spiking to 542 W. A PSU sized on
TDP arithmetic can pass every average and still hit overcurrent shutdown
on a spike.
- Where the numbers come from. — Measured
figures were read from the cited TechPowerUp reviews' own power charts
(via Internet Archive snapshots of those exact pages, recorded per
source). A few CPUs no read chart covers are budgeted at their official
Intel PL2 / AMD PPT power limit instead — those say "power-limit budget"
wherever they appear, never pretending to be measurements.
- Sustained load — CPU measured heavy
all-core peak + GPU measured gaming draw + a fixed 75 W allowance for
motherboard, RAM, drives, fans and USB. Budgeting the CPU at its all-core
peak is deliberate: games draw less, but renders, encodes and shader
compilation hit it.
- Headroom — sustained × 1.15. Margin
for capacitor aging, case heat, and staying near the efficiency curve's
comfortable region.
- Spike check — CPU peak + the GPU's
measured transient spike + 75 W. The recommended size covers this total
outright, so even a pre-ATX-3.0 unit cannot be spike-tripped. ATX 3.0/3.1
units are additionally required to ride through excursions up to 200% of
their rating, which is why some verdicts say "yes on ATX 3.x, risky on
older units."
- Recommendation — the smallest standard
size (450 / 550 / 650 / 750 / 850 / 1000 / 1200 / 1500 W) covering both
the headroom figure and the spike total. The reasoning is always shown —
never a bare number.
- Honesty — one number per part per
metric, each with its citation on the dataset page.
Different board partners' cards and different test scenes measure
differently; these are representative review figures, not guarantees for
your specific card.