A solar server's power budget, device by device

I started the solar server project in July with a spreadsheet instead of a shopping cart. The rule was simple: know what every device pulls before deciding how much panel and battery it needs.

That order matters more than it sounds. Solar gear is sold by the watt, and it is easy to buy a panel because it was a good deal, then a battery because it looked about right, and only later discover the two don’t match the thing you actually want to run. Working backwards from the load is slower on day one and much cheaper by day ninety. It is also the most honest way to see where the energy goes, which is half the point of building anything off-grid.

The always-on loads

These are the machines that run around the clock. They set the floor for everything else.

“Per day” is just watts × 24 hours. The OptiPlex alone is 960 of those 1,060 watt-hours. Everything else on the list together uses about a tenth of what it does.

That one line changes the whole design. If the OptiPlex could be replaced by a single-board computer for most of its jobs, the daily load would drop from about a kilowatt-hour to about a hundred watt-hours, and the panel and battery needed would shrink by roughly the same factor. For comparison, the solar edition of Low-tech Magazine runs on a small single-board server that draws between 1 and 2.5 watts depending on traffic. A reused desktop is a fine place to start, because the most sustainable computer is usually the one already on the bench, but it is worth knowing exactly what it costs to keep it.

The sometimes loads

An EliteBook laptop at about 30 W while it’s in use, plus USB lights, phone charging, a soldering iron and a couple of fans. These come and go, so I budget them by hours instead of by day: four hours on the laptop is another 120 Wh.

A simple way to keep these honest is a small table of device, watts while running, and hours per typical day, then the same multiplication as above. Anything with a heating element (the soldering iron, a kettle, a space heater) deserves a hard look, because heat is where small systems get eaten alive.

The supply side

The array is planned around eight used 300 W panels, 2.4 kW on paper. A cautious planning figure for the Carolinas is about 4.5 peak sun hours a day averaged over the year, and real systems lose roughly a quarter of that to heat, wiring, dust and charging:

2,400 W × 4.5 h × 0.75 ≈ 8,100 Wh on an average day

That’s several times the load, which is the point. Winter days are short, and a week of rain happens.

“Peak sun hours” trips a lot of people up. It isn’t the number of hours the sun is up. It is the day’s total sunlight squeezed into the equivalent number of hours at full strength (1,000 watts per square metre). A long, hazy summer day and a short, crisp winter one can land at very different numbers even with similar daylight. NREL’s free PVWatts calculator will estimate this for your exact address, tilt and panel direction, month by month, which is a better planning number than any regional average. The month that matters is the worst one, usually December or January, not the yearly mean.

The 0.75 factor is also a planning figure, not a law. Panels lose output as they heat up, wires lose a little over every metre, charge controllers and inverters have their own losses, and pollen and dust settle on the glass. A quarter is a reasonable allowance for a small, carefully wired system; a long cable run or a cheap inverter can push it higher.

The battery

The battery has to carry the night, plus a cushion for cloudy days. Fourteen hours of darkness at 44 W is about 620 Wh. Two fully overcast days on top of that is roughly 2,100 Wh more, so about 2.7 kWh usable. Lithium iron phosphate batteries are comfortable down to about 20% charge, which puts the bank at around 3.4 kWh.

Two choices hide inside that paragraph.

How many cloudy days to plan for. Two is a judgment call. More days means a bigger, more expensive battery that sits mostly full in summer. Fewer days means accepting that the server sometimes goes dark in a long grey spell. Websites like Low-tech Magazine’s solar edition make the second choice on purpose: they aim for about 90% uptime, which works out to the site being offline around 35 days a year, and they say so on the page. For a server that holds a community’s maps and files, a bigger cushion makes sense. For a mirror of a website that is also hosted elsewhere, a smaller one is fine.

Which chemistry. Lithium iron phosphate (LiFePO4) costs more up front than lead-acid, but it tolerates deeper discharge, lasts many more cycles, and doesn’t need to be kept topped up to avoid damage. Over the life of a system it is usually the cheaper and less wasteful choice, and it contains no cobalt.

What the numbers say to do first

1. Shrink the big load. Moving the OptiPlex’s always-on jobs onto the Le Potato, and waking the desktop only when it’s needed, cuts the battery requirement more than any purchase could.
2. Measure before buying. A plug-in power meter costs less than a single panel and turns every number in this entry from an estimate into a fact.
3. Plan around the worst month. Size the array for winter and enjoy the surplus in summer. The surplus is a feature: it can charge tool batteries, run a fan, or be shared.

Next

Measure each device with a plug-in power meter. Averages are a starting point, not a promise.

Further reading

• PVWatts Calculator from the U.S. National Renewable Energy Laboratory. Free month-by-month solar production estimates for any address.
• How to build a low-tech website, Low-tech Magazine. The hardware, power draw and uptime trade-offs behind a solar-powered web server.
• I put a server in my shed, Read The Manual, October 2026. An honest build log where an 11 W Raspberry Pi load (about 276 Wh a day) outran a 35 W panel, and what the rebuild changes.
• OpenEnergyMonitor. Open-source hardware and software for monitoring electricity, solar and storage.
• Three projects from one pile of hardware, the projects this budget is for.