Ops & troubleshooting / 2026-10-08
Three Physical Constraints of a Small Server Room: Power, Heat, Noise
Three machines in one rack: performance and stability hit power and cooling limits first. These numbers matter more than the CPU model.
In a self-hosted setup, machines rarely fail on "not enough compute" first — more often power and heat hit their limits first. Three numbers are worth working out up front.
1. Headroom on a single circuit
A normal wall outlet is 10A per circuit, which at 220V is about 2200W, but sustained load above 80% is not advisable, so around 1700W.
A dual-socket Xeon with a few cards pulls 600–900W at full load without much effort. Two machines like that plus a switch and a NAS and you're at the limit. The symptom is a tripped breaker, or a hot plug.
2. Heat density in the rack
In a small 12U rack, 25℃ intake and 40℃ exhaust is a common state. If the intake-to-exhaust delta exceeds 15℃, airflow is insufficient and heat is circulating inside the rack.
A practical check: put your hand at the exhaust — "hot but tolerable" (around 50℃) is still normal; if it's hot enough that you pull your hand back instantly, the airflow path has a problem.
3. Noise and fan RPM
A 1U server's fans ramp to high speed at 40℃ intake, and noise can exceed 60dB. Put that in a bedroom or living room and the end result is usually "switched off and never used".
Three workable directions:
- Move to a tower case with large, slow-spinning fans
- Pick a low-power platform (Xeon E series, Ryzen + consumer board)
- Put the high-load machines on the balcony or in a storage room, and keep only the network gear indoors
A mistake I made
During the two hottest weeks of summer, CPU temperature went from 62℃ to 78℃. The machine didn't break, but two mechanical drives hit 54℃ and reallocated sectors started showing up in SMART.
Conclusion: fix the drive temperatures first, worry about the CPU later. Drives are far more fragile than CPUs.
A minimal checklist
- [ ] Full-load power < 80% of the circuit
- [ ] Intake-to-exhaust delta < 15℃
- [ ] Mechanical drive temperature < 45℃
- [ ] Independent temperature/humidity monitoring (even a cheap sensor)
- [ ] UPS on power loss, or at least a safe shutdown
The three physical constraints of a small server room
- Heat: a single 1U server at a sustained 200W can push a closed rack past 45℃ in 20 minutes, and the drives give out first.
- Noise: fans are regulated by CPU temperature, so they go to full speed as soon as load picks up — basically impossible to live with at home.
- Power: running 24×365, at 0.6 yuan/kWh, 200W is 1050 yuan a year — more than a lot of cloud VPS plans.
Do the math before you buy
| Item | Estimate |
|---|---|
| Whole-system power draw (incl. drives) | 150–250W |
| Annual consumption | 1300–2200 kWh |
| Annual electricity cost | 800–1300 yuan |
| Extra air-conditioning load | +200–500 yuan depending on the room |
A used server is "cheap" only on the premise that you don't count electricity and noise.
Practical ways to cut noise and heat
- Swap the fans: replace 40mm high-RPM fans with larger low-speed ones — trade temperature for noise.
- Cap the power:
powertop --auto-tuneplus setting PL1 in the BIOS — 10% performance loss for 30% lower power draw. - Separate the drives: put mechanical drives in an external enclosure instead of cramming them into the same airflow path as the motherboard.
One rule of thumb
If the room is above 32℃ for long stretches, or you're sensitive to noise (you can hear the fans at night), don't self-host. Rent a small box in Hong Kong or Singapore — roughly the same cost, zero noise.