A homelab stays on day and night, and that is exactly why power consumption matters more than purchase price. A server drawing a few watts more than another, multiplied by every hour of the year, can cost more on the bill than you saved by buying it used. Before optimising, though, you need to measure.
Why nameplate figures are not enough
The rating on the power supply is the maximum it can deliver, not what the system uses. A 500 W supply can power a small server that spends the day at 25 W. Likewise, the processor's TDP describes a reference thermal dissipation, not the machine's real consumption.
What matters in a homelab is almost always idle consumption, meaning services running but no significant load. That is the state the server spends most of its time in.
Measuring tools
Plug-in power meter
The simplest tool: it sits between the wall socket and the power strip and shows instantaneous power and accumulated energy in kWh. Left connected for a week, it gives a realistic figure that includes night-time backup peaks and quiet periods.
Smart plugs with metering
Some smart plugs measure power and make it available on the network. Integrated into a home automation system or a metrics database, they let you draw historical graphs and compare before and after every change. Prefer models that work locally, without depending on the manufacturer's cloud service.
UPS and server sensors
Many UPS units report load in watts or as a percentage, and servers with a BMC expose power supply consumption over IPMI. They are useful for a quick reading, but accuracy varies: the plug meter remains the reference.
Working out the yearly cost
The formula is simple:
yearly kWh = average watts × 24 × 365 / 1000
yearly cost = yearly kWh × price per kWh
An example with round numbers: a system averaging 40 W uses about 350 kWh a year. At a hypothetical price of 0.30 euros per kWh, that costs about 105 euros. The same calculation for an old 120 W workstation gives more than 1,000 kWh. Take the energy price from your own bill, using the total cost per kWh rather than just the energy component.
Compare that number with what you saved at purchase: often a more expensive but frugal mini PC pays for itself within a few years.
Where the watts hide
Always-on components
- Spinning hard drives: every disk that spins draws a few watts even when idle. Six drives weigh on the total.
- Add-in cards: RAID controllers, 10 gigabit network cards and dedicated GPUs have noticeable idle draw and can stop the CPU from entering its deepest power-saving states.
- Fans and oversized power supplies: a power supply is less efficient when delivering a small fraction of its rated output.
Processor power states
Modern processors cut consumption through C-states (idle states) and frequency management. If they are disabled in the BIOS, perhaps by a «performance» profile, the system uses more than it needs to. On Linux, the powertop tool shows how long the CPU spends in each state and suggests tweaks, such as power management for PCIe and USB devices.
Practical steps to reduce consumption
- Check the BIOS: enable C-states, ASPM for the PCIe bus and balanced power profiles.
- Consolidate: two lightly loaded old servers can often become a single machine running several virtual machines.
- Separate archive and services: rarely read data can sit on drives that spin down after a period of inactivity, while always-on services run on SSDs.
- Switch off what is not needed all the time: a backup server can power on only at night, with scheduled power-on or Wake-on-LAN.
- Remove unneeded hardware: a graphics card left in the slot only for installation keeps drawing power.
- Measure after every change: without a comparison you do not know whether it worked.
A note on spinning drives down: stopping and restarting them often increases mechanical wear. It makes sense for archive drives that sit idle for hours, not for volumes accessed every few minutes.
Tracking over time
In my work I record homelab consumption alongside other metrics: CPU, temperature, disk load. When a software change, such as a service that never lets the drives sleep, raises average consumption, the graph shows it at once. A simple sheet with date, change and measured watts is already enough to see which interventions are worth it.
In short
- What counts is average idle consumption, not the power supply rating or the TDP.
- A plug meter left connected for a week gives the most reliable figure.
- Yearly cost comes from average watts, hours in a year and the real price per kWh.
- Spinning drives, add-in cards and disabled C-states are the most common hidden loads.
- Measure every optimisation before and after.
