The number that fails is not the one people calculate
A wattage sum answers a steady-state question. Every part in the machine is drawing something, you add those somethings up, and the total is what the supply has to deliver on average. That arithmetic is straightforward and it is almost always fine. Builds that add up to 500 watts and run on a 750 watt supply do not fall over because of arithmetic.
What does cause failures is the shape of the draw over time. A modern graphics card can go from an idle figure to several times its rated draw and back inside a millisecond or two, faster than any voltage regulation loop reacts and far faster than the average power meter shows. If that excursion is large enough, the supply's over-current or over-power protection sees a fault and does what it was designed to do, which is shut everything off. The machine reboots, the log says nothing useful, and the wattage sum still says the supply is adequate — because for the average it is.
This calculator gives you both numbers, and it makes the spike multiplier an input rather than asserting one, because the honest answer is that the multiplier belongs to a specific part and not to a category. Where you get it: the manufacturer's own documentation sometimes states a recommended supply that is far above the card's rated draw, and that gap is the spike allowance in disguise. Independent measurement with a scope is the other source. A number you found in a forum thread about a different card is not.
Where the wattage figures should come from
Every wattage in the default list is a placeholder chosen to make the form do something on first load. None of them describe a part you can buy, and they will drift out of date the moment a generation turns over. Replace them.
| Item | Where the real figure lives |
|---|---|
| Graphics card | Board power or total graphics power in the manufacturer's specification page, not the review headline |
| Processor | The package power figure, plus whatever the specification calls the short-duration or boost limit if it states one |
| Motherboard and onboard devices | Rarely published. A measured wall figure at idle with a card removed is more honest than a guess. |
| Drives | Datasheet active figures. Spin-up on a mechanical drive is well above its running draw and lasts seconds, not milliseconds. |
| Fans and pumps | Current rating on the label times the fan voltage |
| Anything over USB | The port budget rather than the device, if the device does not say |
Where a figure genuinely cannot be found, put in something generous and mark the line so you remember it is a guess. A wattage budget with one honest unknown in it is more useful than one with nine confident inventions.
Choosing the headroom figure
The headroom field exists because there is no correct percentage and anyone who tells you there is has stopped thinking. What the margin actually buys is four separate things, and how much of each you want is yours to decide: room for parts you have not bought yet, cover for wattage figures you are not certain of, distance from the point where the supply's fan spins loudly, and distance from the point where an ageing supply that has lost some capacity stops coping.
Running a supply near its rating is not inherently harmful — supplies are rated to deliver their rating. It is simply that everything about the experience gets worse as you approach it. Efficiency curves usually sag at both extremes, fan curves ramp, and the margin for a component you add in two years disappears. A number somewhere between twenty and fifty percent is where most people land, and the reason to prefer the top of that range is future parts rather than safety.
Efficiency, the wall, and the electricity bill
The supply's efficiency does not change what the components need. It changes what the socket delivers to get there, and it decides how much heat ends up inside your room rather than inside the machine. A build drawing 500 watts of DC through a supply running at 90 percent efficiency pulls about 556 watts from the wall, and the 56 watt difference leaves as heat from the supply itself.
That figure feeds two other calculations worth doing. The running cost is the obvious one, and if the machine is on for long stretches the electricity bill calculator and the appliance running cost calculator handle it against your real tariff. The less obvious one is that every watt from the wall becomes heat somewhere, which is what the case airflow calculator takes as its input. A quiet machine and an efficient one are more closely related than they look.
What this deliberately does not do
It does not tell you which supply to buy, name an efficiency certification as a requirement, or claim a wattage for any part. It also does not model per-rail limits, which matter on some designs and not at all on others, or connector counts, which are a compatibility question rather than a power one. Check both against the specific supply's own documentation.
And it will not help you diagnose a supply that is already misbehaving. If a machine shuts off under load, the useful next steps are testing with a different supply and checking whether the shutdowns correlate with a specific workload, not recalculating the sum. The arithmetic was probably right the first time. Related sizing sits in the power supply rail calculator for low-voltage electronics work and the circuit load calculator if the question has moved from the machine to the breaker feeding it.
Questions people ask
Why do supply recommendations sit so far above the calculated draw?
Because they are covering the transient behaviour, not the average. A manufacturer that recommends a supply several hundred watts above the card's rated draw is buying margin against the millisecond spikes and against the possibility that your supply is older, hotter, or of a design with tight protection thresholds. They are also covering themselves against every possible rest-of-system, since they have no idea what else is in your machine. Treat that recommendation as one input alongside your own sum rather than as a competing answer.
Is a bigger supply less efficient at low load?
Usually a little, at the very bottom of the curve. Efficiency curves typically peak somewhere in the middle of the load range and fall off at both ends, so a very large supply running a very small load sits in the less efficient region. The effect is real but small in absolute watts — a few percent of a small number. It is not a good reason to buy a supply with no headroom, and the actual shape of the curve is in the supply's own documentation rather than in a general rule.
Does the spike multiplier apply to everything at once?
No, and treating it that way overestimates. The calculator sums each line at its own multiplier, which is the pessimistic case where every part peaks in the same instant. In practice the spikes from different components are uncorrelated and rarely coincide, so the true worst case is somewhere between the continuous total and the summed peak. Leave the multiplier at 1 for anything that does not have a documented transient behaviour, which is most of the list.
Can I measure the draw instead of adding it up?
Yes, and it is better where you can. A plug-in wall meter gives you the real figure at the socket for a machine you already own, which you divide by the efficiency to get the DC side. Two caveats. A cheap meter averages over a window long enough to hide exactly the transient behaviour that matters, so it tells you about the continuous number and nothing about the spike. And it can only measure what is already built, which does not help when the point of the exercise is deciding what to buy.
What about the mains side and the outlet?
The current figure at the bottom of the results is what the machine draws from the socket. For a single desktop on a general-purpose circuit this is rarely close to anything, but it stops being trivial when a room holds several machines, monitors, and a space heater on one breaker. That is a branch circuit question rather than a supply question — the circuit load calculator covers it. Anything involving the mains wiring itself, including anything inside the supply enclosure, is not user-serviceable work.