By LK Wood IV · 2026-06-04 · ~11 min read · St. Louis County, MO
The question “how much does a homelab cost to run?” gets hand-waved more often than it gets answered. This article answers it with specific hardware numbers, not vague estimates.
Idle power — what hardware draws when powered on and not under active workload — is the baseline cost you pay 24 hours a day, 365 days a year. Everything else (peak load, backup windows, encode runs) adds on top. For 24/7 homelab hardware, idle power is the majority of your annual electricity cost.
Methodology
Power figures below come from three sources, in order of preference:
- Measured values — published in third-party reviews that explicitly state Kill-A-Watt or at-the-wall measurements (ServeTheHome, STH forums, AnandTech archives, specific community hardware testing threads cited below)
- Manufacturer specifications — published limits provide context where a wall-power test is unavailable; a CPU rating does not establish complete-system idle draw
- Platform-class estimates — for generic device categories where no specific hardware is identified
Figures marked with (M) are measured values from cited sources. Figures marked with (E) are estimates based on methodology above. One caveat decides how several rows below are labelled: a vendor spec sheet publishes maximum draw, not idle, so a spec sheet on its own can never establish an idle figure. Rows whose only source is a spec sheet are marked (E) and say what the vendor actually publishes. Use the Power & Cost Calculator to plug in your specific hardware — these are starting points.
Mini PC / SFF nodes (Proxmox hosts)
Use a named review configuration to budget for a mini PC. The same processor can appear in different machines with different memory, drives, firmware and network interfaces. The following published measurements were rechecked October 4, 2026; they remain observations of the configurations tested on their original review dates.
| Tested system | Published idle measurement | Configuration and source |
|---|---|---|
| Beelink EQ14 / Intel N150 | 8.91W one-minute average (M) | Notebookcheck, January 27, 2025; 16GB RAM, 512GB SATA SSD, dual gigabit Ethernet, Windows 11 Home; power test and configuration |
| GMKtec NucBox G3 Plus / Intel N150 | 9.3W at the wall (M) | Jeff Geerling’s January 2025 Ubuntu 24.04 test record; replacement SSD, with cooling modifications and High Performance mode documented in the record; test notes |
| Minisforum MS-01 / Core i9-13900H | 25–29W (M) | ServeTheHome, January 9, 2024; 32GB/1TB review configuration and power measurement |
The review conditions differ. Do not rank the two N150 boxes by their fractional-watt difference or transfer the MS-01 figure to every CPU/storage configuration. For the first pair, our EQ14 versus G3 Plus comparison explains which ports, drive slots and memory limits change the buying decision.
Download the mini-PC power reference CSV. It retains the publication date, checked date, source URL and test context beside every row for use in your own comparison spreadsheet.
Work out the annual cost before buying
For a 365-day year, annual kWh = average watts × 8.76. Multiply by the variable electricity rate on your bill. A flat $0.20/kWh is an illustrative rate in this table, not a current national average:
| Constant average load, an example rather than a product prediction | Annual energy | Annual cost at $0.20/kWh |
|---|---|---|
| 10W | 87.6kWh | $17.52 |
| 15W | 131.4kWh | $26.28 |
| 30W | 262.8kWh | $52.56 |
An extra 5W at that rate is $8.76 per year. Compare that with the delivered price difference, the RAM you need and any expansion hardware. A lower purchase price can outweigh several years of a small power-cost difference. Use the Power & Cost Calculator to substitute your tariff, operating hours and expected active load.
For Proxmox, budget the guests first with the capacity planner. Then use the mini-PC buying guide to select the storage and network configuration that carries them. A low idle measurement is valuable only if the computer does the required work.
Tower servers and workstations
| Device | CPU | Idle Power | Source |
|---|---|---|---|
| Dell PowerEdge R720 (1P, base) | Xeon E5-2620 | 80–100W (E) | Platform-class estimate; no measurement of this configuration is linked |
| Dell PowerEdge R730 (1P) | Xeon E5-2600 v4 | 90–120W (E) | Platform-class estimate; no measurement of this configuration is linked |
| Dell PowerEdge R720 (2P loaded) | 2× Xeon + 16 DIMMs | 150–200W (E) | Platform-class estimate; no measurement of this configuration is linked |
| Custom tower, Ryzen 7 7800X3D + RTX 5080 | Ryzen 7 7800X3D / RTX 5080 | 80–100W (M) | My own system, Kill-A-Watt at wall |
| Custom tower, Ryzen 5 7600X, no GPU | Ryzen 5 7600X | 45–65W (E) | AMD platform characteristics |
| HP EliteDesk 800 G5 SFF | Core i7-9700 | 20–30W (M) | STH community, HP published specs |
Takeaway: Enterprise rack servers from 2013–2018 have high idle floors. A 2U Dell R720 with a single CPU idles at an estimated 80–100W — 5–10× a modern mini PC for comparable compute on lightly-threaded workloads. They shine for large RAM configurations and enterprise expansion cards at low per-GB RAM cost. For power-conscious homelabs, used SFF business PCs (HP EliteDesk, Dell OptiPlex) offer a middle ground.
NAS devices
| Device | Config | Idle Power | Source |
|---|---|---|---|
| Synology DS923+ (no drives) | Ryzen R1600 platform | 20–30W (M) | Synology spec sheet, community measurements |
| Synology DS923+ (4× HDD, spinning) | + 4 × 4TB HDD | 35–50W (E) | NAS + drive idle additive |
| TrueNAS Scale node (mini PC) | N100 + 4-bay HBA | 20–30W (E) | Mini PC idle + HBA + drives |
| QNAP TS-464 (no drives) | N5105 | Under 21.6W (E) | Estimated under QNAP’s published 21.6W disk-standby figure, which is measured with all four bays populated; QNAP publishes no diskless idle |
| Terramaster F4-424 Pro (no drives) | N100 | 15–20W (E) | N100 platform |
| DIY NAS (used HP Microserver G10+) | AMD 200GE | 30–45W (E) | AM4 platform estimate |
HDD idle note. A spinning 3.5" HDD uses 4–6W at idle (heads unparked, spinning). In standby (spin-down), it drops to 0.5–1.5W. NAS units with spin-down configured can save 15–25W if drives are mostly idle. ZFS is not friendly to spin-down (frequent scrubs and cache writes keep drives spinning) — factor this in if you’re running TrueNAS.
NVMe vs HDD for NAS power. Replacing four 3.5" HDDs with four NVMe drives saves approximately 12–20W at idle. NVMe idles at 0.5–2W per drive. At $0.184/kWh, 15W savings over a year = 131 kWh = $24/year. At California rates (~$0.333/kWh), $44/year. For a 24/7 NAS, the power argument for NVMe is real alongside the performance argument.
Networking
| Device | Type | Idle Power | Source |
|---|---|---|---|
| MikroTik hEX (RB750Gr3) | 5-port 1GbE router | 5–7W (E) | Estimated from MikroTik’s published 5W without attachments / 10W max |
| MikroTik CRS305-1G-4S+IN | 4-port 10GbE SFP+ | 10W idle, 13W max observed (M) | ServeTheHome’s review (“Idle Power: 10W, Max Observed Power: 13W”), the figure our 10GbE guide uses; MikroTik publishes 10W without modules, 18W max |
| MikroTik CRS326-24G-2S+RM | 24-port 1GbE + 2× 10GbE | 18–22W (E) | Estimated below MikroTik’s published 24W max; no idle figure published |
| Ubiquiti UniFi US-8-60W (PoE) | 8-port PoE switch | 10–16W (E), not counting PoE draw | Estimated from Ubiquiti’s published specs; no idle figure published |
| Cisco SG350-28 | 28-port managed | 35–50W (E) | Derived from the Cisco data sheet; cisco.com returns 403 to automated re-checks, so not re-verified 2026-08-22 |
| TP-Link TL-SG108E | 8-port unmanaged 1GbE | 3–5W (M) | Multiple reviews |
| Netgear GS308E | 8-port managed 1GbE | Under 3.7W (E) | Netgear publishes 3.7W maximum consumption and no idle figure |
Takeaway: At 1GbE the small fanless switches all land within a couple of watts of each other. Netgear rates the GS308E at 3.7W maximum and the TP-Link TL-SG108E measures 3–5W. Pick on ports and features. MikroTik earns the efficiency crown once you need 10GbE, and the CRS305 gives you 4× 10GbE SFP+ for 10W idle in ServeTheHome’s measurement — remarkable for what it does. Cisco enterprise switches at 35–50W for 28 ports are fine in an office with hundreds of devices; for a homelab with 8–12 devices, they’re wasteful.
Annual cost by configuration
Using $0.184/kWh (US residential average, EIA May 2026), 24/7 operation:
| Configuration | Total Idle W | Annual kWh | Annual Cost |
|---|---|---|---|
| Starter (N100 mini PC + 4-bay NAS) | 28–40W | 245–350 kWh | $45–64 |
| Standard (N150 mini PC + 4-bay NAS + 8-port switch) | 35–55W | 307–482 kWh | $56–89 |
| 3-node cluster (3× mini PC + NAS + 10GbE switch) | 70–120W | 613–1,051 kWh | $113–193 |
| Single used rack server (R720 1P) | 80–100W | 701–876 kWh | $129–161 |
| GPU workstation AI node (RTX 5080, idle) | 85–100W | 745–876 kWh | $137–161 |
| Full setup (3-node + NAS + GPU workstation + switch) | 155–220W | 1,358–1,927 kWh | $250–355 |
The mini PC vs rack server gap. A 3-node mini PC cluster idles at 60–90W total. A single 2U Dell R720 idles at an estimated 80–100W. The three mini PCs offer more total cores, better per-core performance, less rack noise, and lower electricity cost than the single enterprise server. For homelab use cases, the argument for rack servers rests on RAM capacity (DDR3 RDIMM is cheap), storage slots, and expansion cards — not power efficiency.
Regional rate comparison
At the same 40W idle draw (standard single-node setup), annual electricity cost by US region:
| Region | Avg Rate | Annual Cost |
|---|---|---|
| Missouri | ~$0.137/kWh | $48 |
| Louisiana | ~$0.142/kWh | $50 |
| Texas (ERCOT) | ~$0.164/kWh | $58 |
| US Average | ~$0.184/kWh | $64 |
| New York | ~$0.299/kWh | $105 |
| California (SCE/PG&E average) | ~$0.333/kWh | $117 |
| Hawaii | ~$0.52/kWh | $182 |
Sources: EIA Electric Power Monthly, May 2026 data. Residential average rates including all fees.
California and Hawaii homelab operators pay 2.3–3.8× more per kWh than Missouri and Louisiana. At 40W idle (350 kWh/year), that’s an extra $67–134/year. That is meaningful but not prohibitive. At 200W idle (1,752 kWh/year, full multi-node setup), that gap becomes $335–671/year, which starts to inform hardware efficiency decisions.
Reducing idle power
Power profiles. Set the host CPU governor to powersave or schedutil on Proxmox:
echo powersave | tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
On always-on nodes with modest VM workloads, powersave governor can cut idle draw by 10–20% with no visible performance impact.
Drive spin-down. On NAS units, configure drives to spin down after 20–30 minutes of inactivity. This saves 15–25W when the NAS is sitting idle overnight. ZFS users: disable spin-down or configure a scrub schedule that doesn’t run overnight when you want spin-down to activate.
Shut down what you’re not using. A mini PC you’re not actively using that sits at idle 22 hours/day is 22 hours of electricity for zero benefit. Schedule shutdowns or use Wake-on-LAN to bring nodes up only when needed.
Efficient NAS sizing. A 2-bay NAS at 20W beats a 4-bay NAS at 40W if you only need 2 drives of storage. Size the enclosure to what you need, not what you might grow into.
Use the Power & Cost Calculator to model your specific setup — enter each device’s measured or estimated idle watts and see the annual cost. The calculator also shows the vs-cloud delta, which contextualizes the power cost against what you’d pay for equivalent hosted infrastructure.
For UPS sizing based on your total idle draw, use the UPS Runtime Calculator to model runtime at your specific load. For mini PC picks that optimize the efficiency numbers above, see Best Mini PCs for the Homelab 2026.
Frequently asked questions
How much does a typical homelab add to an electricity bill?
Is a mini PC really that much more efficient than a tower server?
Do hard drives contribute significantly to idle power?
Does a managed switch use much power?
How accurate are manufacturer TDP specs for idle?
Sources and corrections
- Last updated
- Methodology
- See our methodology for research and review standards.
- Update log
11 dated updates are retained in the full log.
Read the full update log (11 entries)
- 2026-10-04 — Mini-PC section rebuilt around three attributed, configuration-specific published tests: EQ14, G3 Plus and MS-01. Added a source/context CSV, explicit illustrative cost math and a product comparison. Removed the generic TDP percentage method; existing owner tower measurements are unchanged.
- 2026-09-28 — The tower table lists my own Ryzen 7 7800X3D and RTX 5080 tower at 80-100W at the wall, read on a Kill-A-Watt. The annual cost table carries the GPU workstation and full-setup rows, and the first FAQ answer gives the GPU node’s 85-100W.
- 2026-09-28 — Rack-server rows relabelled. The single-CPU R720, single-CPU R730 and loaded dual-CPU R720 rows were marked (M), measured, while citing unlinked ‘STH community’ and ‘STH forum reports’ sources. No linkable measurement of those configurations was found (the ServeTheHome forum thread ‘Dell r730 vs r720 power usage’ has one owner’s R730xd with one CPU and ten drives at about 125W, not the base configuration), so under the page’s own legend they are (E) and say so. The figures are unchanged. The two sentences that repeat the R720 figure now say it is an estimate and give the R720 as 2U, which Dell’s R720 technical guide states (‘Form factor 2U rack’); they said 1U. The mini PC FAQ gave an N150 box 15-22W and the R720 80-120W, neither of which is in the table; it now carries the table’s 10-15W and the estimated 80-100W.
- 2026-09-28 — MS-01 row corrected. It read Core i9-12900H, 30-45W (M), citing an unlinked ServeTheHome review. ServeTheHome’s review power page reads ‘We saw around 25-29W at idle on our test system’, and page 1 names the unit ‘a top-end configuration with an Intel Core i9-13900H’ (both fetched 2026-09-26). The row now carries the i9-13900H, 25-29W and the link, matching the mini PC guide. The mini PC takeaway moved with it: ‘single-digit watts’ and ‘25-45W’ became 6-15W and 20-35W, the ranges the table’s own rows give.
- 2026-09-24 — CRS305 row, FAQ and takeaway corrected. The page carried 6.5-7.1W idle citing our 10GbE guide, which no longer says that: its 2026-09-24 pass replaced the unsourced figure with ServeTheHome’s measurement (Idle Power 10W, Max Observed Power 13W, spec sheet 18W). This page now carries the same measured figures and the source.
- 2026-08-23 — Provenance correction. A site audit on 2026-08-23 found that the 2026-08-22 edit added the legend rule that a vendor spec sheet publishes maximum draw and so can never establish an idle figure, then left two rows marked (M) whose only source was a spec sheet. Netgear GS308E: netgear.com/business/wired/switches/plus/gs308e/ publishes “Power Max consumption (Watts)” = “3.7W maximum” (fetched 2026-08-23), so the row’s 5-8W (M) sat above the vendor’s own maximum; it now reads under 3.7W (E). QNAP TS-464: qnap.com returns HTTP 202 with an empty body to automated fetches, so I read the spec from the Wayback capture of the vendor page (web.archive.org/web/20251125230827/https://www.qnap.com/en-us/product/ts-464/specs/hardware), which publishes “Power Consumption: Disk Standby Mode 21.618 W” and “Operating Mode, Typical 40.536 W”, both “Tested with drives fully populated”. No diskless idle figure exists, so the row is now (E) and says so. The Ubiquiti US-8-60W row carried no mark at all and is now (E) on the same rule. The switch FAQ repeated the overstated Netgear band and the networking takeaway claimed MikroTik was the most efficient managed switch outright; I corrected both to match the table.
- 2026-08-22 — The opening FAQ said the MikroTik CRS305 draws 7-9W while the networking table on the same page carries the third-party measured 6.5-7.1W, and this site’s CRS305 review carries the same 6.5-7.1W. I corrected the FAQ to the measured figure.
- 2026-08-22 — Provenance correction on the networking table. Four rows were marked (M) - which this page defines as ‘measured values from cited sources’ - while citing only a vendor spec sheet, and a spec sheet publishes maximum draw, not idle. Verified directly at mikrotik.com 2026-08-22: hEX RB750Gr3 10W max / 5W without attachments, CRS305 18W max / 10W without attachments, CRS326 24W max, none publishing an idle figure. The three spec-derived rows are now (E) and state what the vendor actually publishes; the Cisco row is (E) and flagged as not re-verifiable (403). The CRS305 idle figure itself was 7-9W here, 6.5-7.1W on our own review page and 8-10W on the 10GbE guide; all three now carry the third-party measured 6.5-7.1W bare figure. Legend extended to say why a spec sheet alone cannot establish idle.
- 2026-08-14 — Corrected two cost figures that contradicted this article’s own tables. (1) The opening FAQ claimed a starter homelab adds ‘$7-10/month’ and a 3-node cluster ‘$15-20/month’; the Annual cost by configuration table on the same page gives 28-40W ($32-46/year) and 70-120W ($80-137/year), which are $3-4/month and $7-11/month. The FAQ overstated running cost by roughly 2.5x. Re-derived from the table at 8,760 h/year and the article’s own $0.13/kWh assumption. (2) The regional paragraph said California and Hawaii cost ‘an extra $55-91/year’ more than Louisiana and Missouri at 40W, but $55-91 is the delta against Texas ($0.12/kWh), not against Louisiana ($0.09) or Missouri ($0.11). Against the two states actually named the delta is $60-102/year at 40W and $298-508/year at 200W, and the rate multiple is 2.5-4.2x rather than 3-4x.
- 2026-08-19 — Re-derived every cost figure that used the $0.13/kWh US-average rate after the site’s canonical rate moved to $0.184/kWh (EIA Electric Power Monthly, May 2026 US residential average). Wattages and kWh totals are unchanged — only the rate and its derived dollars moved: the configuration table’s $32-46 / $40-63 / $80-137 / $91-114 / $97-114 / $177-251 became $45-64 / $56-89 / $113-193 / $129-161 / $137-161 / $250-355, the opening FAQ’s annual and monthly figures moved with it, the NVMe-vs-HDD example (15W saved, 131 kWh) went from $17/year to $24/year, and the small-switch FAQ line was re-based at the same 5W floor to roughly $8/year. Each old figure was first reproduced from its stated wattage at 8,760 h/year and $0.13 before substitution. Per-state rows and figures citing their own state rates (including the regional table, already on May 2026 data) were not touched. The hero SVG and its alt text carried the same table and were updated to match.
- 2026-08-22 — On 2026-08-22 I corrected the NVMe-vs-HDD example. It said ‘At California rates, $37/year’ for 131 kWh, which reproduces only at the retired ~$0.28/kWh California figure (37 / 131.4 = $0.282/kWh). This page’s own regional table gives California ~$0.333/kWh (EIA Table 5.6.A, May 2026: 33.25 cents/kWh), so 131.4 kWh x $0.3325 = $43.69, now $44/year with the rate shown inline. I re-checked everything else at $0.184/kWh and 8,760 h, and it reproduces: the configuration table’s kWh and dollars (28-40 W = 245-350 kWh = $45-64 through 155-220 W = 1,358-1,927 kWh = $250-355), the FAQ monthly figures, the 5 W switch line ($8), and the regional table against EIA 5.6.A May 2026 residential cents/kWh (US 18.44, MO 13.68, LA 14.15, TX 16.44, NY 29.93, CA 33.25, HI 52.00) with the 40 W and 200 W deltas ($67-134 and $335-671). The EIA table is still labelled ‘May 2026 and 2025’.
- Corrections
- Spotted an error or a stale number? Email contact@techfuelhq.com. Confirmed corrections are recorded here with the date and change.