I run a small home lab. It hums in the corner of my office, a collection of single board computers, a mini PC, a network switch, and an external hard drive. It hosts my personal websites, a media server, a development environment, and a handful of cron jobs that keep my digital life organized. I had always assumed it was cheap to run because the individual components are low power. But I never actually measured it. Assumptions about power consumption are like assumptions about subscription fees: small things add up invisibly. So I bought a smart plug with energy monitoring, plugged my entire lab into it, and tracked every watt for a full month. The numbers surprised me, and what I learned changed how I build and run my home infrastructure.
The Gear Under Test
My home lab is not a rack of enterprise servers. It is a modest collection of devices that a hobbyist accumulates over years of tinkering. The main workhorse is a used Lenovo ThinkCentre M720q Tiny PC with an Intel Core i5-8500T processor, 32GB of RAM, and a 512GB NVMe SSD. It runs Proxmox as a hypervisor with several virtual machines and containers: Home Assistant, a PostgreSQL database, a Node.js web server, and a Plex Media Server that streams to my TV. I also have two Raspberry Pi 4 units. One runs Pi-hole and a WireGuard VPN server. The other is a dedicated OctoPrint server for my 3D printer, though the printer itself has a separate power supply and was not included in this measurement. A Netgear 8-port unmanaged gigabit switch connects everything. An external 4TB Western Digital USB hard drive hangs off the ThinkCentre for media storage and backups. My internet router and modem are on a separate outlet and were not part of this test. All of these devices, except the printer, were plugged into a single power strip fed through a Kasa KP115 smart plug that tracks energy usage in real time.
The Measurement Setup and a Surprising Initial Reading
The Kasa plug reports current power draw in watts and cumulative energy use in kilowatt hours. It updates every few seconds and logs data to the Kasa app, though I exported the daily totals to a spreadsheet for analysis. On the first day I plugged everything in and watched the real-time reading settle. I expected maybe 20 watts at idle. The actual number was 48 watts. I thought the plug was faulty. I disconnected the hard drive and the Pis, leaving only the ThinkCentre. The reading dropped to 32 watts. The two Raspberry Pis together drew 7 watts. The switch added 3 watts. The hard drive idled at 5 watts and spun up to about 8 watts during file transfers. The total idle draw of the entire lab was 48 watts continuous, day and night. That is a lightbulb’s worth of power, burning 24 hours a day. I had never thought of my lab as a lightbulb that never turns off.
The Real World Usage Patterns
Over the month, I logged the daily kilowatt hour readings and noted what the lab was doing on high-activity days. Most days, the lab idled with occasional CPU spikes when cron jobs ran or when I SSHed in to do some work. The ThinkCentre’s power management was set to balanced, meaning the CPU scaled down to around 900MHz at idle and ramped up to 3.2GHz under load. During a typical idle hour, the lab drew between 46 and 50 watts. When I ran a larger task, the power draw jumped noticeably. Transcoding a video for Plex pushed the ThinkCentre’s CPU to near 100 percent and the total lab draw spiked to 68 watts for the duration of the transcode. Running database migrations and rebuilding Docker containers pushed it to 62 watts. Even a modest task like updating all packages on the virtual machines raised the draw to 55 watts for about ten minutes. These spikes were brief, but they added up.
The biggest continuous consumer was the external hard drive. Even at idle, it drew 5 watts because it never spun down. The Proxmox host accessed it frequently for VM backups and media library scans, so the drive was always active. I had not considered the constant power cost of a spinning disk in a setup that otherwise used efficient ARM and mobile Intel chips. The hard drive alone consumed 3.6 kilowatt hours per month, which was nearly as much as both Raspberry Pis combined.
Calculating the Monthly and Annual Cost
I pay an average of $0.14 per kilowatt hour, which is roughly the U.S. national average for residential electricity. The smart plug reported that the lab consumed exactly 35.4 kilowatt hours over 30 days. That works out to an average of 1.18 kilowatt hours per day. Multiply by my rate and the monthly cost was $4.96. I rounded up to five dollars. Annually, the lab costs about $60 to run. Sixty dollars a year is not a crisis. But it was about twenty percent higher than my gut estimate. Before measuring, I had guessed the lab cost around $4 a month, maybe less. I was off by a dollar a month, which is twelve dollars a year, which would buy a decent pizza. That is not life changing, but it adjusted my mental model of what “low power” actually means. A Raspberry Pi sips power, but a cluster of devices adds up. The ThinkCentre alone accounted for about two thirds of the total consumption. The Pis, the switch, and the external drive made up the rest.
I also calculated the cost of leaving the lab on continuously versus turning it off for eight hours each night while I slept. The lab draws about 48 watts on average over a day. Turning it off for eight hours would save about 0.38 kilowatt hours per day, or 11.5 kilowatt hours per month. That is about $1.61 per month in savings, or $19 per year. The convenience of always-on services, like Pi-hole and the VPN, is worth $1.61 a month to me. But the math showed that if I had a larger lab, the idle power savings could become meaningful.
The Pi-hole and VPN on a Pi: The Efficiency Champion
I measured the two Raspberry Pi 4 units individually by temporarily plugging only them into the smart plug for a few hours each. The Pi running Pi-hole and WireGuard drew a steady 3.5 watts with the official power supply. The OctoPrint Pi drew 3.8 watts because it occasionally activated the printer’s camera and had a slightly higher CPU load. Together they used 7.3 watts. Over a month, that is about 5.3 kilowatt hours, costing about $0.74. That is genuinely negligible. If I moved the VPN and Pi-hole to a dedicated low-power device like a Pi Zero 2 W, which draws under 1 watt, I could cut that already tiny number further. But the effort required to migrate the services is not worth the $0.50 monthly savings. The Pi 4 is efficient enough for always-on services, and the headroom is valuable.
The Hard Drive Idle Drain and the Spindown Fix
The external hard drive was the least efficient part of the lab. A 3.5-inch spinning disk in a USB enclosure, it used 5 watts at idle and 8 watts under load. The disk never spun down because Proxmox accessed it every few minutes for a backup check. I investigated configuring spindown using hdparm, but the USB bridge in the enclosure did not reliably support the command. The drive would spin down and immediately spin back up, which is worse for the drive and uses more power than just staying on. The correct solution would be to replace it with a USB SSD. A 1TB SATA SSD in a USB 3.0 enclosure idles at under 1 watt. That swap would save about 4 watts continuously, or 2.9 kilowatt hours per month, about $0.41. The cost of the SSD would not be recovered in electricity savings alone, but the reliability and noise benefits would make it worthwhile. I added that upgrade to my list for the next hardware refresh.
The Network Switch and the Phantom Loads
The Netgear gigabit switch used 3 watts continuously. That is negligible, but I was curious about whether a managed switch with energy-efficient Ethernet would make a difference. My research indicated that energy-efficient Ethernet can reduce port power when cables are shorter or idle, but the savings are a fraction of a watt. For an 8-port switch, the difference is maybe 0.5 watts. Not worth replacing a perfectly functional switch. The lesson was that network gear is already very efficient, and obsessing over a watt or two is a distraction from bigger opportunities.
I also discovered a phantom load I had not considered: the power strip itself had a tiny LED indicator and some basic surge protection circuitry. It drew 0.3 watts. That is essentially nothing. But it reminded me that every powered device, even a power strip, contributes to the baseline. The sum of many tiny loads is what creates the difference between my estimate and reality.
What Surprised Me Most
The biggest surprise was not the total cost. It was the proportion of power used by the ThinkCentre compared to the dedicated SBCs. A single mini PC with a 35-watt TDP processor, even at low utilization, dominates the energy budget of a small lab. The Raspberry Pis are efficient but not powerful enough for many server tasks. The tradeoff between capability and power is real. Running a hypervisor with multiple VMs lets me do more with fewer physical devices, but the base power draw is higher than running several dedicated SBCs. For my workload, the ThinkCentre was the right compromise. For someone who only needs a file server and Pi-hole, a single Raspberry Pi or an Intel N100 mini PC with a 6-watt TDP would be far more efficient. I am now more aware of matching the hardware to the actual workload, not just picking the most capable device available.
I was also surprised by how consistent the power draw was. Apart from short spikes, the lab hummed along at a nearly flat 48 watts. Modern CPUs with good power scaling and SSDs that idle efficiently keep the baseline predictable. That made it easy to estimate monthly costs from a single day’s measurement. I had expected more variability from background tasks, but the automation scripts and cron jobs were so lightweight that they barely moved the needle.
What I Changed After the Experiment
Armed with real data, I made a few changes. I consolidated two VMs that were doing similar background tasks into a single container, which reduced the average CPU load slightly and saved maybe 2 watts. That is tiny, but it also simplified management. I configured the Plex server to transcode to RAM instead of disk, which reduced drive writes and might lower the drive’s active time a bit. I replaced the external hard drive with a 1TB SSD in a USB enclosure, which dropped the idle draw from 5 watts to 0.8 watts. That swap also made backups faster and quieter. The total idle draw of the lab is now 43 watts, down from 48. That saves about $0.70 per month, or $8.40 per year. The SSD cost $60, so the payback period is about seven years from electricity alone. The real benefit was the noise reduction and the reliability improvement. Sometimes the numbers justify the purchase; sometimes the comfort does.
What I’d Do Differently
If I were starting a home lab from scratch with power consumption as a priority, I would choose different hardware. I would use a single low-power mini PC with an Intel N100 processor, which idles at around 6 watts and can still run multiple containers and VMs. I would skip the external spinning drive entirely and rely on an internal NVMe SSD with a USB SSD for backups. I would use a single Raspberry Pi for the always-on network services. The entire lab would idle at about 10 to 12 watts instead of 43. That would cost around $1.20 per month instead of $5, saving over $45 a year. The upfront cost might be slightly higher for a new N100 mini PC, but the long term savings and reduced heat output would be worth it for a build meant to run 24/7 for years.
I would also measure first, then build. I spent years guessing that my lab was cheap to run. A $15 smart plug gave me certainty in a month. If I had measured earlier, I would have caught the hard drive idle drain sooner and made more informed hardware choices. The smart plug has paid for itself in knowledge, if not directly in electricity savings.
The Bigger Picture for Home Lab Owners
My home lab costs $5 per month to power. That is less than a streaming subscription. It is not a financial burden, and it is not the place to find meaningful household savings. The real value of measuring is understanding your infrastructure. When you know exactly how much power each component uses, you make better decisions about upgrades and architectures. You also develop an intuition for the energy cost of always-on computing, which is useful when designing systems for others or when considering whether to leave a machine running overnight.
If you run a home lab, spend the fifteen dollars on a smart plug and do your own measurement. Your numbers will differ from mine. Your lab might be a rack of old enterprise gear that pulls 200 watts, or a single Raspberry Pi that pulls 3. Either way, knowing is better than guessing. The data will either reassure you that your setup is efficient, or it will point you toward the largest energy consumer in the stack. In my case, it was the mini PC and the spinning hard drive, and now I know exactly what to optimize when the time comes.
