PSU calculator that prints the bill of watts

Two separate reasons, and neither is a safety fudge factor. The first is that the certification on the box is measured at 20, 50 and 100 percent of rated load, and every tier records its best figure at the middle one — sizing so your peak lands near 70 percent puts everyday use, which is far below peak, on the part of the curve that was actually tested. The second is that a graphics card is allowed to draw multiples of its rated power for microseconds at a time, and the margin that absorbs those excursions has to come from somewhere. A supply run at its label does both jobs badly.

You are almost certainly watching a one-second average of a fault that lasts a millisecond. Software monitoring polls at intervals far longer than a transient, so a card that briefly demands twice its board power shows up as nothing at all in the log, while the supply's over-current protection — which reacts in microseconds — sees it and latches off. A multi-rail unit makes this worse, because the limit that trips is the one on a single 12 V rail rather than the total, so a build well inside the overall rating can still overload one connector.

An older unit works, but it needs to be bigger to do the same job. The ATX 3.0 design guide, published in 2022, was the first to define the excursion envelope explicitly and to require the supply to ride it out at its rating, so a compliant unit already contains the margin. A supply designed before that has no such rating, which is why the second figure on this page targets 55 percent load instead of 70 — you buy the missing margin as extra watts. ATX 3.1 in 2024 kept the envelope and tightened the connector rules around it.

It is five percentage points of efficiency at half load — 90 percent certified against 85 — which on a machine drawing 450 W is about 30 W of heat that never enters the room, or 44 kWh over a year of four hours a day at that load. The electricity alone therefore repays the price difference slowly. The stronger arguments are the ones that do not appear in the certification: less waste heat means a slower fan and a quieter machine, and tiers above Bronze are usually built on newer topologies with better hold-up time and tighter voltage regulation. Buy the tier for the build quality it correlates with, not for the arithmetic.

Because it carries up to 600 W through one plug, which leaves no room for a bad connection. The original 12VHPWR design would happily deliver full power through a partly seated plug, concentrating the current into whichever pins were making contact and melting them. The 12V-2x6 revision that came with ATX 3.1 shortens the four sense pins so that a plug which is not fully home is detected and limited to 150 W instead. Whichever version your card uses, push until it clicks and check that no gap is visible at the shroud before you close the case.

Barely, and much less than the folklore suggests. Efficiency does fall away below 20 percent of rated load, so a 1000 W unit idling at 60 W is working outside its tested range, but modern supplies hold 80 to 85 percent even there and the absolute loss is a handful of watts. The real cost of oversizing is not the meter, it is the money spent on capacity you never use and the semi-passive fan sitting stopped so long that dust settles inside. Two sizes above the recommendation is waste; one is insurance.

About 9 W while it is reading or writing, but roughly 24 W on the 12 V rail for the two or three seconds it takes to spin the platters up. That surge is what catches a build with several disks that all start at once when the power button is pressed, and it is why server backplanes stagger their spin-up. The steady figure is the one in the table here, since a supply sized for a graphics card has plenty of margin for a couple of drives; if you are filling eight bays, look for staggered spin-up in the firmware rather than for more watts.