Power, Noise, and Thermals

Choosing Hardware touched on power draw and noise as constraints worth weighing before buying. This page goes a level deeper, because unlike a one-time purchase price, these are costs that compound for as long as the machine stays plugged in.

Power: measure it, don't guess it

Manufacturer TDP figures describe peak draw under load, not what a machine actually pulls sitting mostly idle running a few containers - the number you actually care about. Measure the real thing with a cheap plug-in power meter (a "Kill A Watt" or a smart plug like a Shelly Plug that reports wattage over the network):

Idle draw:        12W   (mini PC, Docker running, low traffic)
Typical draw:      18W   (a couple of services doing real work)
Peak (transcoding): 35W  (Jellyfin transcoding one 1080p stream)

Turn that into an actual cost:

watts ÷ 1000 × hours running × price per kWh = cost

18W ÷ 1000 × 24 hours × 365 days × $0.16/kWh ≈ $25/year

A modern mini PC idling around 10-20W costs roughly $15-30/year to run continuously at typical US residential electricity rates - genuinely cheap. A used rack server idling at 100-150W is $140-210/year for the same continuous-uptime pattern, before it's doing anything. That gap is the real-world version of the "used enterprise gear is cheap upfront, expensive to run" trade-off from Choosing Hardware.

Noise: the constraint that's easy to underrate on paper

Decibel ratings on spec sheets are measured in ideal conditions and rarely match lived experience, and fan noise character matters as much as volume - a low, constant hum is far more tolerable long-term than a fan that revs unpredictably under load. Two practical rules:

• If it needs to live somewhere people sleep, work, or relax, prioritize fanless or large-slow-fan designs over compact designs that rely on small, fast (louder) fans to move the same air.
• Test before you commit, if you can - a quiet showroom or a YouTube review doesn't tell you how a specific unit sounds under sustained load in your actual room, with your actual ambient noise floor.

If you already own a loud machine and moving it isn't an option, undervolting the CPU (trading a bit of peak performance for meaningfully lower heat and fan speed) is worth investigating before assuming you're stuck with the noise.

Thermals: the constraint that shows up as a problem later, not upfront

Heat mismanagement doesn't usually fail loudly - it shows up as throttled performance or, over years, shortened component lifespan. Worth checking on any always-on machine, especially one tucked into a closet or cabinet:

• Airflow, not just ambient temperature. A machine in a closed cabinet with no airflow will run hotter than the same machine in an open room at the same ambient temperature - heat needs somewhere to go, not just space to sit in.
• Dust filters, if the case has them. Dust buildup on intake fans and heatsinks is a slow, silent efficiency killer - a five-minute compressed-air clean every few months keeps thermal headroom intact.
• Drive temperatures, especially for spinning HDDs, which have a narrower safe operating range than most SSDs. Check with:

sudo apt install smartmontools
sudo smartctl -A /dev/sda | grep Temperature

• Sustained load, not idle, is the real test. A machine that's perfectly cool at idle can throttle noticeably during something sustained like a Jellyfin transcode (see First Services) or a large backup job. If you plan to run either regularly, check temps during one, not just at rest.

The honest bottom line

None of this needs to be solved perfectly before you continue - it needs to be known. A loud, hot, power-hungry machine tucked somewhere nobody minds is a perfectly fine homelab. The failure mode this page is trying to prevent is discovering the noise or the electricity bill three months in, after everything's already wired up and the returns window has closed.

Next: Choosing an OS.