Noise is the homelab problem nobody warns you about until the rack is already in the bedroom. The spec sheets quote "fanless" for the gateway, "quiet" for the switch, and nothing at all for the mini-PC, which means the buyer learns about acoustic floor the same way every homelabber does: in the dark, at 2 AM, listening to a 40 mm fan spin up because a PoE camera negotiated. This post is the long answer to a short question — how do you take a stack of UniFi, HP, and Lenovo gear and get the rack below the threshold where you stop noticing it. It is also the long version of the case for spending an hour with a screwdriver before spending another evening trying to sleep through a switch.
The short answer, for readers who only have a minute: the room noise floor is usually 28–32 dB(A), the goal for a bedroom rack is to stay inside about 5 dB of that, and most stock fans inside UniFi and small-form PCs are running 8–15 dB(A) above the floor. A handful of Noctua substitutions and a couple of decoupling tricks bring the whole rack back into the "you have to listen for it" range. The rest of the post is the measurement method, the per-device numbers, and the gotchas the YouTube videos do not cover.
How dB(A) actually works in a small room
Decibels are logarithmic, which is the source of every misunderstanding in this discussion. A 3 dB increase corresponds to a doubling of acoustic power. A 10 dB increase corresponds to roughly a doubling of perceived loudness, which is not the same thing as a doubling of energy. The A-weighting on dB(A) is a filter applied to the raw SPL reading that approximates the frequency sensitivity of the human ear — we hear 1–4 kHz much more clearly than 50 Hz, and A-weighting bakes that in.
For a homelab in a bedroom, the meaningful targets are a function of the ambient noise floor in that room. A quiet suburban bedroom at night sits around 28–32 dB(A); a city apartment with a window cracked is usually 35–40 dB(A); a basement with a furnace cycling is 40–45 dB(A) when the furnace runs. The rack disappears acoustically when it is within roughly 3 dB of the room floor and becomes noticeable around 6–8 dB above it. Above 10 dB the rack is the loudest thing in the room.
An SPL meter on a phone is accurate enough to make decisions. Take the reading at 1 m from the rack, at ear height, with the room silent and the rack the only source. Note both the average and the peak; fans that idle quiet and spin up loud are honest at the average and obnoxious at the peak.
The UniFi switches
UniFi switches are the loudest single category in most homelabs, and the noise profile depends heavily on which model is in the rack.
The USW-Lite-8 PoE is fanless. It also runs warm. Measured at 1 m, a stock Lite-8 PoE adds approximately 0 dB to the room floor — meaning, you cannot tell it is on with the cover closed. That is the gold standard for a bedroom rack, and it is the reason this switch sells in the volumes it does despite costing more per port than its 16-port sibling.
The USW-Lite-16 PoE has two small fans. Stock, at 1 m, with light PoE load (one access point), the measured noise sits around 38–40 dB(A). Push the PoE budget toward its rated 45 W with two or three powered devices and the fans ramp up; the noise climbs into the 44–48 dB(A) range and stays there. The fans themselves are 30 mm units that spin near 6,000 RPM at maximum, which produces a tonal whine that the ear locks onto. Replacing the stock fans with Noctua NF-A4x10 or NF-A4x20 PWM units is straightforward, but the switch reads fan tach over a 4-pin header and will alert if it sees zero RPM. The Noctua units run at 4,500–5,000 RPM peak rather than 6,000, so they report tach to the controller cleanly and bring the measured noise down to about 33–35 dB(A) at full PoE load.
The USW-Pro-24 PoE is a different animal. Three 40 mm fans, a real PoE budget, and an enclosure designed around airflow rather than around silence. Stock noise at 1 m is in the 50–55 dB(A) range on idle and climbs to 60+ dB(A) under load. There is no realistic Noctua mod that brings the Pro-24 PoE into bedroom territory; the switch wants air it cannot get from a quiet fan. The mod that works is moving the switch into a closet or another room and accepting that this is not a bedroom switch. If the switch has to stay in the bedroom, the path is to replace the fans with NF-A4x20 PWM units, accept that the switch will likely throttle PoE under sustained heavy load, and undervolt the fans through the existing PWM signal until the thermals are at the edge of acceptable. Even with that mod the floor is around 42–45 dB(A).
The USW-Pro-XG-8 PoE deserves a separate paragraph. It is 10 Gigabit, PoE++, and built around a heat budget that small fans cannot dissipate quietly. Stock noise hovers around 48–52 dB(A) at idle. A Noctua mod helps marginally — about 4–5 dB — but the switch is fundamentally not a bedroom device. Plan for it to live in a closet or a separate room with a fiber run back.
HP EliteDesk and ProDesk Mini PCs
HP's G3 through G5 Mini PCs — EliteDesk and ProDesk — share a 65 mm-tall chassis with a single blower-style fan exhausting out the back. Stock idle noise at 1 m is approximately 30–33 dB(A), which is on the edge of audible in a quiet bedroom and inaudible in any other room. The problem is the fan curve under sustained load; a G4 or G5 running a Proxmox node with three or four small VMs can sit in the 38–42 dB(A) range continuously, with brief spikes higher when something compiles or backs up.
The fan in the HP Mini is a custom blower, not a standard 40 mm or 60 mm unit, so a direct Noctua swap is not realistic. The mods that work are softer:
- Repaste the CPU. HP's factory paste degrades fast on these units, and a fresh application of Arctic MX-4 commonly drops sustained load temperature 8–12°C, which is 2–4 dB at the ear.
- Update BIOS. HP has shipped fan-curve tweaks as late as 2022 and 2023 firmware that meaningfully slow the ramp on workloads that historically triggered it.
- Cap the CPU package power in BIOS. Limiting long-term power to 25–30 W on a 35 W chip costs 5–10% on sustained workloads and drops average noise by 4–6 dB.
- Replace the thermal pad on the SSD — the M.2 sits over the airflow path, and a hot drive triggers fan ramp independently of the CPU.
A fully tuned HP Mini drops to about 28–32 dB(A) idle and 34–37 dB(A) under typical sustained load. That is well inside bedroom territory.
Lenovo ThinkCentre Tiny
The Lenovo M710q, M720q, M920q, and M10q chassis share a similar 65 mm enclosure with a blower fan, but Lenovo has historically tuned the fan curve more aggressively on the quiet side. Stock idle at 1 m is 28–31 dB(A) on most units — effectively the room floor in a typical bedroom. Under sustained Proxmox load with two or three VMs, a stock M920q sits around 34–37 dB(A), which is meaningfully quieter than the equivalent HP G5 out of the box.
The same mods that help an HP Mini help a ThinkCentre Tiny: repaste, BIOS update, and a CPU power cap. The package-power cap on Lenovo is in BIOS as "Intel SpeedStep" and related items, and on later models in the ThinkCentre BIOS menu under "Performance." Capping a 35 W TDP M920q to a 25 W sustained limit drops average noise from around 35 dB(A) under load to around 31 dB(A), which is the difference between "you can hear it if you listen" and "you cannot."
The ThinkCentre fan, like HP's, is not a standard size. A Noctua replacement is not the right tool here; the BIOS work is.
Noctua substitutions and the gotchas
Noctua fans are the default upgrade for a reason: they are quiet at any given RPM, they have honest specs, and the PWM curves they ship are well-behaved. For rack work, the relevant models are the NF-A4x10 (40×10 mm, 5,000 RPM), the NF-A4x20 (40×20 mm, 5,000 RPM, more static pressure), the NF-A6x25 (60 mm), and the NF-A8 (80 mm).
Three gotchas come up repeatedly. First, fan tach. Most switches and many small-form PCs read fan tach and will alert or run the remaining fans at 100% if any fan reports zero RPM. Noctua fans report tach, but the reported RPM can sit below the firmware's expected floor. The fix is sometimes a firmware tach threshold change, sometimes a pull-up resistor on the tach line, more often picking a Noctua whose minimum RPM is above what the firmware expects.
Second, voltage. Noctua's low-noise adapters are inline resistors that drop 12 V to 7 V or 5 V, trading RPM for silence. They work on devices whose firmware does not enforce a minimum RPM, and trigger alerts on devices that do.
Third, static pressure. A switch fan moves air through a dense fin array, not across an open chassis. A Noctua replacement with lower static pressure may be quieter at the same RPM but lose more pressure across the fins, which means the device runs hotter for the same RPM. The static-pressure-optimised models — the "A" series ending in 20 rather than 10 — are the right pick for switch-fan replacements.
Decoupling, fan curves, and where the rack actually sits
Fan substitutions are the headline mod, but three other tricks add up to roughly the same noise reduction with less screwdriver time.
Decouple the rack from the room. A small wall-mount cabinet sitting flush against drywall transmits fan vibration into the wall, which radiates as low-frequency hum throughout the room. Four 25 mm rubber feet between the cabinet and the wall, or four neoprene washers between the cabinet and its mounting bracket, kill most of the structure-borne noise. The change is on the order of 2–4 dB at the ear but the perceptual change is larger because low-frequency hum is something the ear locks onto.
Tune fan curves where the firmware allows. UniFi switches do not expose fan curves, but UniFi's newer firmware adds a "quiet" option in the device settings on a few models that biases the curve toward higher temperature setpoints in exchange for lower RPM. Use it where it is offered. For mini-PCs, the BIOS-level fan curve is the main lever, and on most Lenovo and HP chassis the "Quiet" profile is the right pick over "Balanced."
Move the rack. The single most effective noise mod is to put the rack on the other side of a wall. A small wall-mount cabinet in a closet or laundry room, with a 1 m fiber run back to the bedroom for whatever needs to be in the bedroom, reduces noise at the ear by 15–20 dB compared to having the same rack in the room. This is not a mod; it is the avoidance of the problem. For homelabbers in apartments where there is no other room, the closet trick — rack inside a closet with louvered or vented doors — gets most of the benefit.
A target stack for a bedroom rack
For a homelabber building from scratch with bedroom-noise constraints, the stack that hits a measured 32–35 dB(A) at 1 m under typical load is roughly:
- Fanless gateway (UniFi Cloud Gateway Ultra or Express).
- USW-Lite-8 PoE for downlink, fanless.
- A ThinkCentre M720q or M920q with BIOS power cap as the compute node.
- A small fanless NAS or a single-disk box on rubber feet.
- The whole rack on neoprene washers, with cable bundles tied off the fan exhausts.
Adding the USW-Lite-16 PoE in place of the Lite-8 with a Noctua mod pushes the floor to around 35–37 dB(A), which is still acceptable for most rooms. Adding the Pro XG 8 PoE or Pro 24 PoE forces the rack into a closet.
Wrap-up
A quiet rack is not one product choice; it is a stack of small decisions that add up to a measurable difference in dB(A). The biggest single lever is whether the loud devices — the Pro-class switches, the high-density PoE units — are in the room at all. The second is the fan in the switch that has to be there. The third is the small mods on the mini-PCs that bring them from "noticeable" to "you have to listen for it."
If you remember one number, make it the room-floor target. The bedroom does not need a silent rack; it needs one within 5 dB of the floor already there. Most stock UniFi-and-mini-PC stacks miss that by 10–15 dB. An hour with a screwdriver and a couple of Noctua fans usually closes the gap.
