PSU efficiency at low load is the weakest part of a switching supply’s performance curve, and an idling desktop sits right in it. Certification tiers are awarded on measurements taken at moderate and heavy loads, where conversion is at its best. Below those points the curve bends downward, sometimes sharply, and no badge on the box warns you.
Why the curve sags at the bottom
A switching supply has fixed overheads that do not shrink when the load does. The controller chip needs power. The standby converter that keeps the motherboard awake for a wake signal runs constantly. Gate drive losses in the switching transistors depend on switching frequency rather than on output current. Magnetic components have core losses that are largely independent of load. At heavy load these overheads are a small fraction of the total and efficiency looks excellent. At very light load they can rival the useful output itself, and the ratio of useful watts to total watts collapses. The physics is the same in a phone charger and a data center rectifier; only the numbers differ. Standby conversion is the least visible of these, running whenever the machine is plugged in so that a wake signal can be received, and it continues even when the computer is switched off.
Where a real desktop actually spends its hours
Add up how a typical machine is used. Reading, typing, browsing, video calls, and background updates all sit far below the supply’s rating. Even a gaming machine spends more hours at the desktop than in a demanding title. If the supply was chosen with generous headroom, the idle operating point can fall to a small single-digit percentage of rated output, well below the lightest load point any certification tier requires. This is the practical case against extreme oversizing that has nothing to do with whether the machine is stable: a supply spending its life below its measured range is running at an efficiency nobody has published and the owner cannot look up. Multiple monitors compound this, since each display has its own fixed overhead and several idle displays can exceed the idle draw of the computer driving them.
What newer certification tiers changed
Certification programs eventually acknowledged the problem by adding a lighter measurement point for their upper tiers, so the best-rated units are held to a standard at a fraction of the load where older tiers said nothing. This is a genuine improvement and it is one of the few concrete reasons to prefer an upper tier for a machine that idles a lot. Note that the requirement applies to the tiers that specify it, not to every certified unit, so check which tier a candidate unit holds and whether that tier includes a light-load requirement. A manufacturer that publishes its own efficiency curve across the full range is telling you more than the badge does. Manufacturers who publish a full efficiency curve rather than three points are giving you the information the tier omits, and that curve is worth asking for.
How much energy this actually costs you
Keep the arithmetic honest. The absolute waste at idle is small in watts, because the load itself is small. Poor efficiency on a low load still produces a low number. What makes it worth thinking about is duration: a machine left on around the clock multiplies a small figure by a large number of hours. The correct response is usually not a different supply but better power management, because a computer that sleeps properly draws a tiny fraction of what it draws while idling awake. Enabling sleep, shortening idle timeouts, and disabling unnecessary wake sources will beat any efficiency tier upgrade by a wide margin. Wake-on-network settings are the usual culprit for a machine that never stays asleep, and disabling them for devices that do not need them often fixes the problem outright.
Practical steps that actually move the number
Size the supply so that typical load lands in its efficient band rather than clinging to the bottom of its range. Enable the operating system’s sleep and display timeouts, and check which devices are permitted to wake the machine, since a network adapter waking the system repeatedly overnight erases the benefit. Unplug or switch off peripherals that draw standby power through the machine. If you run a home server that genuinely cannot sleep, choose the supply on its published light-load efficiency rather than its headline tier, and consider whether the workload could run on a much smaller machine. Measuring the whole machine at the wall with a plug-in meter will tell you where you actually stand. Smart plugs that log energy over time are a convenient way to measure this without leaving a meter in place, provided their own standby draw is accounted for.
What to check before you act
- Measure the machine at the wall while idling, not only under load.
- Enable sleep timeouts and review which devices can wake the system.
- Prefer a tier that includes a light-load efficiency requirement.
- Avoid sizing a supply so large that idle falls off its published curve.
Common questions
Does a smaller power supply idle more efficiently?
Often yes, because the same idle draw represents a larger fraction of its rating and therefore sits higher on the efficiency curve.
Is sleep better than shutting down for energy use?
Sleep uses far less than idling awake and restores instantly. Full shutdown uses the least, but the difference between sleep and off is small compared with the difference between sleep and idle.
The practical takeaway
Fix idle consumption with power management first, then size the supply so normal use lands inside its efficient band. Chasing a higher certification tier while leaving a machine awake around the clock spends money on the smaller of the two problems.
References and further reading
- ENERGY STAR: standby and low-power mode consumption highlights
- Department of Energy: energy-efficient computers and office equipment
- Department of Energy: what your electronics cost to run
Numerical scenarios are illustrative unless identified otherwise. Follow the exact product instructions; component ratings and local installation requirements can differ.
