What Your Home Server Costs to Leave Running
Researched
Seeded with published specifications and cited sources rather than results measured here. Every seeded figure links to where it came from.
Idle accounts for 40% of the yearly bill. That is the part that runs whether you use the machine or not.
This calculator needs JavaScript. The arithmetic is one multiplication chain and it is written out below.
Runs in your browser. Nothing you type here is uploaded, logged or sent anywhere. The page is a static file and the arithmetic happens on your machine. Same principle as the rest of the site.
How this is calculated
Watts to kilowatt hours to money, and nothing else.
idle hours = 24 − load hours
daily kWh = (idle W × idle hours + load W × load hours) ÷ 1000
annual cost = daily kWh × 365.25 × price per kWh
cost per load hour = load W ÷ 1000 × price per kWh
energy per million tokens = cost per load hour ÷ (tokens/sec × 3600 ÷ 1,000,000)
| Constant | Value | Why |
|---|---|---|
| Watts in a kilowatt | 1000 | Meters and tariffs are in kilowatt hours |
| Days in a year | 365.25 | Averages the leap year in rather than drifting a day every four years |
| Seconds in an hour | 3600 | Turns a per-second rate into tokens per hour |
| Idle and load watts | Your measurements | Nothing here is assumed. Measure at the wall |
Worked example, the one above: 40 W for 20 hours is 800 Wh, and 300 W for 4 hours is 1200 Wh. Together that is 2.00 kWh a day, 731 kWh a year, $219.15 at $0.3 per kWh.
What this can't know
- It cannot measure your machine. The watt figures are the whole calculation and they have to come from a plug-in meter at the wall. Power supply ratings are a maximum the unit can deliver, not a draw, and adding up component specifications overestimates badly.
- Two states is a simplification. Real machines sit between idle and full load constantly, spin disks up and down, and boost and throttle. If your workload is bursty, the honest way to use this is to measure an average over a representative day rather than to guess a duty cycle.
- Tariffs move, and some are not flat. A time-of-use or capacity-based tariff cannot be reduced to one number per kWh. Neither can a bill with standing charges, though those do not change when you add a machine, which is why the marginal rate is the right one here.
- Cooling is not counted. Every watt the machine draws ends up as heat in the room. In a climate where that heat has to be removed, the real cost is higher; in a season where it offsets heating, lower. Neither adjustment is in this tool.
- The per-million-token figure is electricity only. It excludes the hardware, which for owned equipment usually dominates. Comparing it directly against an API price is the mistake the break-even calculator exists to stop.
Sources
- The arithmetic is the definition of the kilowatt hour, which is the billing unit for electricity worldwide. No external figure is needed and none is used.
- Your tariff comes from your own bill, and your watt figures from your own meter. That is deliberate: published typical draws for home servers vary so widely that quoting one would be worse than asking.
- No figure on this page came off the bench. When runs are published, this tool gains a machine picker seeded with measured idle and load draws and is re-badged Measured. The method those runs will be held to is on the methodology page.
FAQ
How much does it cost to run a home server all year?
Multiply the average draw in watts by the hours it runs, divide by 1000 for kilowatt hours, and multiply by your tariff. A machine idling at 40 W around the clock and pulling 300 W for 4 hours a day uses about 731 kWh a year, which at $0.3 per kWh is roughly $219. The number that surprises people is the idle line: doing nothing at all costs $105 of that.
Where do I find my electricity tariff?
On your bill, usually as a unit rate per kWh. Use the marginal rate rather than an average that includes standing charges, because the standing charge does not change when you add a server. If you are on a time-of-use tariff and can schedule heavy jobs into the cheap window, use the rate you would actually be paying at the time the machine runs, and be aware that the annual figure then depends on that scheduling holding.
How do I measure idle and load watts?
At the wall, with a plug-in power meter, not from the component specifications. A power supply rating is a maximum it can deliver, not a draw, and adding up part specifications overestimates badly. Read the meter with the machine sitting idle, then again during a sustained job, and give it a few minutes to settle each time. Wall measurements already include power supply losses, which is what you are billed for.
Is the cost per million tokens here the whole cost?
No, and the difference matters. That figure is electricity only. It excludes the price of the hardware, which for owned equipment is usually the larger number by a wide margin, and it excludes network, storage and your time. It is the right figure to compare against the marginal cost of an API call only once you have separately decided the hardware is paid for. The break-even calculator handles the version of this question that includes the purchase.
Why is idle power worth more attention than peak power?
Because it runs for every hour of the year and peak does not. In the example above, 4 hours a day under load contributes less to the bill than 20 hours of doing nothing. Ten watts saved at idle is worth more over a year than a hundred watts saved during a job you run occasionally, which is why the quiet, low-draw machine usually wins on running cost even when it loses on benchmarks.
Where to go next
- The version of this question that includes the price of the hardware: local versus cloud break-even.
- What an always-on disk array adds to the standing draw: the NAS capacity and power calculator.
- Sizing the battery that keeps it alive through a cut: the UPS sizing calculator.