Power Consumption Math: What a 24/7 Mini PC Actually Costs

“How much does it cost to run?” is the first question anyone sane asks before leaving a computer on every hour of every day. The answer is almost always “less than you think” — but the math is worth doing properly, because it changes which hardware decisions actually matter and which are rounding errors.

I run nine low-power machines around the clock and track their draw with an inline meter. This post is the formula I use, the numbers it produces at real electricity rates, and an honest take on when a few watts are worth chasing.

The one formula you need

Annual energy and cost from a constant wattage:

kWh/year = (watts × 24 × 365) / 1000
$/year   = kWh/year × your_rate_per_kWh

A shortcut worth memorizing: at $0.16/kWh (the rough 2026 US residential average), every 1 watt of continuous draw costs about $1.40 per year. So a 10W box ≈ $14/year, a 25W box ≈ $35/year, a 60W box ≈ $84/year. Multiply by your own rate ratio if you pay more or less.

That single sentence — 1 watt ≈ $1.40/year at average US rates — does 90% of the homelab power reasoning you’ll ever need. To run the full calculation against a specific box and your own rate, the Mini-PC Picker and Power-Cost TCO tool computes annual running cost alongside a hardware shortlist.

Use real numbers, not the spec sheet

The wattage you plug into that formula has to be the number at the wall, not the CPU’s rated TDP. A “6W TDP” N100 chip lives in a box that also has RAM, an SSD, a NIC, fans, and a power brick that’s only 80–88% efficient at low load. Real warm-idle draw for that “6W” system is 7–10W.

The only honest way to get your number is to measure it with an inline meter. I cover the full method — cold idle vs warm idle vs 24-hour average, and which one to budget against — in how to actually measure mini PC idle wattage. The short version: budget against warm idle (full stack running, no active load), because that’s what the box does for ~22 hours of every day.

What real boxes cost to run

Using warm-idle figures from my own fleet and the formula above, here’s annual cost at three representative rates. (Rates vary wildly — pick the column nearest yours.)

Machine class	Warm idle	$0.12/kWh	$0.18/kWh	$0.34/kWh
Wyse 5070 thin client	6W	$6	$9	$18
N100 mini PC, full stack	9W	$9	$14	$27
N305 mini PC, full stack	14W	$15	$22	$42
Ryzen 7 mini PC, full stack	22W	$23	$35	$66
Old tower “server” (idle)	70W	$74	$110	$209

• $0.12/kWh ≈ cheaper US regions, parts of the Midwest/South.
• $0.18/kWh ≈ rough US average once delivery and time-of-use are included.
• $0.34/kWh ≈ California peak tiers, much of Western Europe, the UK.

The table makes the real story obvious: the gap between a thin client and an N100 is a few dollars a year and not worth losing sleep over. The gap between any mini PC and a repurposed full-size tower is the entire reason this site exists. Replacing a 70W idle tower with a 9W N100 saves $85–$200 a year depending on where you live — that pays for the mini PC inside a year almost everywhere, and every year after that is profit.

Where chasing watts is worth it (and where it isn’t)

I tune my boxes, but I’m clear-eyed about the payoff.

Worth it:

• Replacing a tower or always-on gaming PC with a mini PC. This is the only change that saves real, three-figure money. Do this first.
• Picking the lower-idle box when two are otherwise equal. Free savings at purchase time — choose the N100 over the N97 if you don’t need the extra clock, choose DDR4 over DDR5 in this class. Roughly $5–10/year, but it costs nothing.
• Killing a dead 70–120W appliance. An old NAS, a forgotten always-on desktop, a power-hungry router. These hide in plain sight and each one can dwarf your whole mini-PC fleet’s draw.

Marginal:

• BIOS tuning (C-states, disabling turbo/radios). Genuinely shaves 1.5–3W and I do it on every box, but that’s ~$3–5/year per machine. Do it because it’s quick and free, not because it’s transformative.
• Swapping a Samsung 980 Pro for a lower-power NVMe drive. Real (0.5–1.5W) but small. Worth doing on a new build, not worth buying a new drive for.

Not worth it:

• Agonizing over thin client vs N100 to save $5/year. Buy the one that fits your workload. The power difference is noise.
• Undervolting an Alder Lake-N chip for idle. The idle gains are negligible because the chip already idles near its floor; you mostly affect load power, which barely happens on a homelab box.

Solar, batteries, and the “it’s basically free” trap

A common claim: “I have solar, so my homelab is free.” Mostly true during the day, but a 24/7 box draws hardest at night when the panels are asleep and you’re either pulling from the grid or draining a battery you paid for. Net-metering rules and battery cycle costs vary too much for a clean rule, but the honest framing is: solar lowers the effective rate, it rarely makes it zero. Run the formula with your true net rate, not $0.

A worked example

My main N100 box: measured 8.6W warm idle, dropped to 7.4W after BIOS tuning. I pay a blended $0.19/kWh on a time-of-use plan.

Before tuning: 8.6W → 75 kWh/yr → $14.30/yr
After tuning:  7.4W → 65 kWh/yr → $12.35/yr
Tuning saved:  ~$2/yr

Two dollars a year from tuning. Meanwhile, that box replaced a Synology that idled near 28W:

Old NAS: 28W → 245 kWh/yr → $46.55/yr
New box:  7.4W → 65 kWh/yr → $12.35/yr
Hardware swap saved: ~$34/yr — every year, forever

The lesson in one line: the hardware class decision is worth 15–20× the tuning. Get the class right, do the easy tuning because it’s free, and stop optimizing after that.

Bottom line

• Memorize: 1 watt ≈ $1.40/year at average US rates. Scale to your rate.
• Budget against measured warm idle, never spec-sheet TDP.
• A whole mini-PC fleet often costs less to run per year than one old idle tower.
• Spend your optimization energy on killing high-watt appliances, not on shaving the last watt off an already-efficient box.

Low-power computing isn’t about obsessing over single watts. It’s about not running a 70W machine to do a 9W job. Once you’ve made that one switch, the rest is rounding error — and you can get back to actually building things. For the build, chip, and storage decisions that set those watts, browse our homelab topic index. For sizing the service stack that drives those watts, SelfhostRealm ↗ is the right starting point, and DockerHomeLab ↗ has the Compose stacks that run lean on hardware this small.