Power & UPS
Home server power budget: idle watts and yearly cost
A NAS that idles at 100W costs about $158/year US, £216 UK, €307 EU. The same workload on a tuned N100 mini-ITX idling at 12W costs $19/year US, £26 UK, €37 EU — a $140-280/year delta that pays for the N100 board in one year. Power optimization on a home server isn't a hobbyist concern in 2026; with energy prices where they are, the wattage difference between a stock untuned tower and a tuned low-power build is the largest single line item in NAS operating cost.
Power adders dominate the budget
Reference images and diagrams. Click any image to view full resolution.
Who this is for
Home operators running an existing NAS or building new in 2025-2026, evaluating 'what does my idle wattage actually cost me per year' and how to reduce it without breaking functionality.
Outcome
A measured idle-wattage baseline, an optimization tactic list ranked by impact, and a 1-2 year cost projection at current electricity rates. Operator can decide whether to keep current setup, tune it, or migrate to a low-power platform.
Required inputs
- A Kill-A-Watt or equivalent power meter — or a smart plug with energy monitoring (TP-Link Kasa KP115, Shelly Plug), or an APC UPS with an LCD that reports load.
- Access to BIOS / firmware settings on the NAS or host.
- Knowledge of current electricity rate ($/kWh or £/kWh or €/kWh).
- Tolerance for downtime to apply BIOS / kernel parameter changes.
Step-by-step procedure
Measure current idle wattage at the wall
Do: Plug NAS power cord into Kill-A-Watt; wait 30 min for HDDs to spin down (if enabled); record idle wattage in two states: HDDs spun down and HDDs spinning.
Expected result: Two numbers: idle-with-spun-down-drives + idle-with-drives-spinning.
If not: Without measurement, you'll optimize blindly. The wattage delta between platforms is too large to estimate.
Calculate annual cost at your local electricity rate
Do: Annual kWh = idle watts × 24 × 365 / 1000; annual cost = kWh × your rate. At the 2026 US average (~$0.18/kWh) every 1 W of 24/7 draw ≈ $1.58/yr: 15 W ≈ $22/yr, 50 W ≈ $74/yr, 100 W ≈ $149/yr, 200 W ≈ $298/yr. In CA / New England (~$0.24/kWh) those roughly double. Compare to a tuned target (~10-15 W for an SSD-only N100; ~25-40 W for a typical Unraid box).
Expected result: A specific $/year figure at your own rate. The gap to target is your optimization budget.
If not: If your current cost is fine (<$50/year), optimization may not justify the work.
Apply the highest-impact optimization first: HDD spin-down (existing systems)
Do: Enable spin-down (Synology: Control Panel > HDD/SSD > HDD Hibernation; Unraid: Settings > Disk Settings; TrueNAS Scale: per-disk). Audit what prevents spin-down: Plex library auto-scan, video preview thumbnails, intro detection — disable all three. Mover schedule on Unraid: change from hourly to nightly. Re-measure after 24 hours of normal use.
Expected result: Drives spun down for 16+ hours/day; idle drops ~4-6W per drive that reaches standby (a 4-bay ≈ 9W saved).
If not: If drives don't stay down, find what's keeping them awake (Plex is the usual culprit). And weigh the tradeoff: spindown suits cold archives, but on an active pool it burns the ~50k start/stop cycle budget and adds spin-up latency — for an intermittent NAS, Wake-on-LAN + full off often saves more than spindown.
Enable PCIe ASPM (systems with dGPU or PCIe-attached NIC)
Do: BIOS > Advanced > PCIe > ASPM = Enabled (L1 or L1 substates). Also enable Package C-State at C10. On Linux: verify with `lspci -vv | grep -i aspm`. On Windows: Power Options > advanced > PCI Express > Link State Power Management = Maximum power savings.
Expected result: Idle wattage drops 5-15W on Intel 12/13th gen; Arc A380 idle drops ~10W; whole system reaches deeper C-states.
If not: If ASPM doesn't reduce power, your motherboard or NIC may not negotiate ASPM cleanly. Realtek RTL8125 2.5 GbE NICs are known to break ASPM on recent kernels — replace with Intel I225/I226 if possible.
Right-size the PSU (new builds or rebuild candidates)
Do: Replace oversized PSU (1000W+) with appropriately-sized Gold/Platinum (450-650W for typical NAS). PSU efficiency tanks below 10% of rated load.
Expected result: 5-15W reduction in measured idle wattage from PSU efficiency improvement alone.
If not: If PSU change doesn't help, your idle wattage may already be in the PSU's efficient range — bigger gains will come from other tactics.
Consider platform migration if delta exceeds 50-80W and savings exceed $100/year
Do: Compare your current idle vs an N100 mini-PC (6-15W) or a tuned i3/i5 build (~10-15W). Example: an R730 at ~200W vs a 15W N100 = 185W delta ≈ 1,621 kWh/yr ≈ ~$295/yr at $0.18 (≈$486 in CA) — a ~$200-300 build pays back in ~8-14 months. Enterprise still makes sense for cheap bulk ECC RAM (relatively attractive in the 2025-26 RAM-price spike), IPMI, many drive bays, or native 10GbE — but avoid DDR3-era gear.
Expected result: Clear migration go/no-go based on cost recovery period.
If not: If your existing NAS is doing 200W idle for a workload an N100 would handle, the migration usually pays back in 1-2 years.
Commands and settings paths
Measure idle wattage at the wall
Plug into Kill-A-Watt; let system idle 30 min; record W reading.
Where: Physical access to the NAS power cord.
Expected: Two numbers: HDDs-spinning + HDDs-spun-down idle wattage.
Failure means: If readings vary wildly, the system isn't truly idle (background tasks running). Disable Plex library scans + Mover before measuring.
Safe next step: Use APC UPS LCD load reading as a substitute if no Kill-A-Watt.
Verify PCIe ASPM negotiated
lspci -vv | grep -i 'aspm'
Where: Linux/Unraid shell.
Expected: Output shows 'ASPM L1 Enabled' or similar on the relevant devices.
Failure means: ASPM disabled = CPU can't reach deep C-states; idle wattage stays high.
Safe next step: Enable ASPM in BIOS; replace ASPM-breaking devices (Realtek 2.5 GbE NICs) with Intel equivalents.
Check CPU C-state residency (the real idle signal)
turbostat --interval 5 (watch Pkg%pc6/pc8 + PkgWatt); `cat /sys/devices/system/cpu/cpuidle/current_driver`
Where: Linux/Unraid/TrueNAS shell.
Expected: Deep package C-state (pc6/pc8) residency >90% at idle, and the driver is `intel_idle` (not `acpi_idle`) on Intel.
Failure means: If the package can't drop below pc2/pc3, one device (often an HBA or non-ASPM NIC) is holding it up — idle stays high.
Safe next step: Enable ASPM/C10 in BIOS; move the boot SSD to a chipset SATA port; identify the offending PCIe device.
Apply and baseline powertop tunables
sudo powertop --auto-tune (persist via a systemd service); baseline with `powertop --time=30 --csv=/tmp/before.csv` then compare after
Where: Linux/Unraid shell.
Expected: System baseline power drops after auto-tune; SATA ALPM + runtime PM show as 'Good' in the Tunables tab.
Failure means: If nothing changes, tunables may already be applied, or a device refuses runtime PM.
Safe next step: Re-run after each BIOS change; keep the systemd unit so tunables survive reboot.
Log watts over time with a smart plug
TP-Link Kasa KP115 or Shelly Plug energy monitoring (Shelly: GET /meter/0 returns power in W)
Where: At the wall, between the NAS and the outlet.
Expected: A continuous W/kWh log; capture idle-spundown, idle-spinning, and load states separately.
Failure means: Wildly swinging reads mean the system isn't truly idle (background tasks) or the plug reports calculated, not measured, watts.
Safe next step: Disable Plex scans / Mover before measuring; prefer a plug that truly measures (Shelly) over one that estimates.
Evidence to record
- Baseline idle wattage measurement (HDDs spinning + HDDs spun down).
- Per-tactic delta: HDD spin-down savings, ASPM savings, PSU swap savings.
- Final idle wattage post-optimization; cost/year math at your rate.
- If considering migration: tentative target platform + cost recovery period.
Common mistakes
- Buying a 'free'/cheap used enterprise server (R730/R710) then paying $250-500/yr in electricity — the ~200W idle dwarfs the purchase savings within a year vs a ~15W N100 build.
- Counting peak/TDP watts instead of idle — the server is at idle ~95% of the time, so idle sets the lifetime bill.
- Adding an oversized PSU 'for headroom' — below ~15% load efficiency collapses (a 750W unit at 30-60W can be ~70%); right-size or use a Pico-PSU.
- Never enabling C-states / ASPM in BIOS — leaves the package stuck at C2/C3, often doubling idle (a single non-ASPM device drags the whole package up).
- Trusting `lspci` 'ASPM Enabled' as proof — it can read enabled while residency stays shallow; verify package C-state residency in powertop/turbostat instead.
- An HBA flashed card idling hot — LSI 9211/9207 draw 6-18W and frequently disable ASPM by firmware default, blocking deep C-states.
- Buying ECC RAM + ECC platform 'because every NAS needs ECC' for media + replaceable dumps — accepts the higher idle cost without the data-criticality justification.
- Counting on HDD spin-down without auditing Plex / Mover / Community Apps for blockers — drives never spin down and the expected savings never materialize.
- Aggressively spinning down an active pool — burns the ~50k start/stop cycle budget, adds latency/timeouts, and in-rush spikes can erase the savings; spindown is for cold archives, not active pools.
- Realtek 2.5 GbE NIC + 'inexplicable' high idle — the NIC breaks ASPM and costs a few watts; replace with Intel I225/I226 for the cleanest path.
- Adding a discrete GPU for transcoding when the iGPU's QuickSync would do — +~35W idle, heat, and noise for zero transcode gain (an N100 iGPU does multiple 4K streams).
- Ignoring your own region's $/kWh — the same 100W box costs ~$87/yr in cheap states but ~$210/yr in CA/New England.
Stop points
- Stop before aggressively spinning down an ACTIVE pool — that's where start/stop cycles, latency, and in-rush spikes bite. Spindown is for cold archives; for an intermittent NAS, Wake-on-LAN + full off often beats spindown.
- Stop before kernel parameter experiments on a production NAS without a known-good rollback (and tested backups). `intel_idle.max_cstate=0` and similar can deeply break a system.
- Stop before buying replacement hardware solely to save power if the cost recovery period exceeds ~2-3 years — and re-check your region's $/kWh, since the payback swings ~2x between cheap and expensive states.
- Stop before adding a discrete GPU 'for transcoding' — confirm the CPU's iGPU QuickSync can't already do it; the dGPU adds ~35W of continuous idle.
- Stop before trusting a software wattage estimate — reconcile against a wall meter (Kill-A-Watt / Kasa KP115 / Shelly), since software ignores PSU conversion loss.
Last reviewed
2026-05-18
Source-backed checks
HomeTechOps turns official docs and conservative safety rules into a shorter runbook. These links are the source trail for the page direction.
Planning a purchase?
We keep a source-backed, price-free comparison so you can buy once and right. No star ratings, every spec cited.
Mini PC for a Home Server: N150 vs MS-01 vs Tiny →