Avalon 1246 – Fan Buzzing

Power cycle at the PDU for 60 seconds. Pull the plug, don't use a web-UI reboot. A full cold start clears any wedged cgminer driver state and lets the fans re-initialize from zero. Note whether the buzz is present on cold start, warm idle, or only at full hash — all three signatures point at different root causes. Roughly `15%` of "buzzing fan" tickets resolve here because the fan controller was stuck at a non-sensical commanded RPM.

Shop-vac the intake grille and chassis exterior. Dust cake on blade leading edges is the #1 cause of buzzing on indoor-deployed A1246s. Vacuum the rear intake grille, the side vents, and any exposed duct surface. Do not point compressed air directly at the fans — blowback can spin them past nameplate RPM and drive back-EMF into the PWM driver IC, creating a new problem you didn't start with. Vacuum pulls, never blows.

Verify ambient + intake clearance. A1246 wants inlet air `≤ 35 °C` (A1066 manual spec, A1246 inherits). Put an IR thermometer or thermocouple at the rear intake grille — not room-middle. Nothing within `15 cm` of the intake. No curtains, dust piles, pet bedding, or cardboard boxes obstructing airflow. Bottlenecked intake forces fans to work harder, accelerates bearing wear, and amplifies any latent imbalance into audible noise. Fix this first; it's free.

Dashboard RPM + `estats` sanity check. Open the Canaan web UI or SSH in and run `cgminer-api -o estats`. Both fans should sit within `±200 RPM` of each other at steady state (~`6000 RPM`). If one fan is `500+ RPM` below its peer, you now have the fan index to target in Tier 2. If both fans look healthy but the buzz persists, the fault is mechanical (shroud / screw / grill), PSU-side, or coil whine — not a bearing problem.

Phone-mic a 10-second audio sample. Drop it into any free spectrum-analyzer app. A steady peak at `60 Hz` or `120 Hz` = transformer / PSU-side. A peak in the `1-6 kHz` band that scales with RPM = bearing or imbalance. Broadband hiss with no fundamental = usually airflow turbulence from an intake obstruction. This step is free, takes 30 seconds, and narrows your Tier 2 work considerably.

Power-off hand-spin each of the two `14038` fans. PDU off, wait 30 seconds. Remove the rear grille (`8× M3×12` screws, Torx T10). Spin each fan firmly by hand and let it coast. Healthy: silent, coasts `6-10 s`. Bad bearing: gritty feel, stops in under `2 s`. Tick per revolution: blade contact or cracked hub. Rank worst-to-best before touching anything.

Inspect each blade under raking light. Remove the suspect fan from its bay (`4× M4` screws per fan). Hold each blade edge against a bright flashlight at a shallow oblique angle. Look for chips, hairline cracks, embedded debris, missing fragments, dust cake. Check the hub for cracks radiating from the centre shaft — a cracked hub fails catastrophically and is a replace-now situation. Check the grill for bent wires and for witness marks where a blade has been rubbing.

Check every chassis and fan screw for back-out. A1246 uses `8× M3` per fan bay plus `4× M4` on each fan mount. Under continuous vibration these can back out `1-2 mm`, which is enough to shift a fan against its duct and produce a tick-per-revolution. Hand-tighten all of them with the correct driver before replacing a fan — do not use power tools, you will strip the chassis threads.

Shop-vac + wet-wipe the grill, duct, and fan cages. Once blades are removed, shop-vac every internal surface a fan sees. Wipe the grill and duct with a lint-free cloth and `99%` isopropyl alcohol. Reassemble. Retest. Many "buzzing fan" tickets resolve here without any fan replacement.

Swap the suspect fan with its peer. Pull both fans. Label slots 1 and 2 with tape. Reinstall them in swapped positions. Power up, let stabilize 10 minutes, re-read `cgminer-api -o estats`. If the fault moves with the fan → bad fan, order a replacement. If the fault stays in the slot → socket / control-board side, skip to Step 13.

Measure socket voltage under load. Suspect fan unplugged, miner hashing, multimeter on DC. Probe `+12V` pin to `GND`: expect `12.0-12.5 V` steady. Probe PWM: should respond to commanded RPM (varies with firmware-requested speed). Probe TACH with a fan plugged in: should show a pulse stream. Socket reads `12.0-12.5 V` steady = socket fine, fan replacement confirmed. Socket reads `0 V` / floats / drifts wildly = Tier 4.

Replace the fan with a matching `14038` PWM unit. Source: Nidec `V35R14BS2M3-07` / `T35R14BS2M3-07`, Sanyo Denki `9GV1412M401`, or equivalent `140×140×38 mm`, `12 V`, `2.4-3.2 A`, 4-wire PWM server fan. Match the connector pinout — Canaan uses the standard `+12V / GND / PWM / TACH` 4-wire Molex PWM header, but confirm against the harness before ordering. Zip-tie the cable away from blade sweep. Reassemble, power up, confirm both fans now read within `±200 RPM` in `estats`.

Check the PSU internal fan. If `PS[0] = 2048` is set, or the buzz is louder at the PSU than at the chassis intake, the PSU fan is your target. Power off, disconnect the PSU, open the PSU case (caps hold charge — discharge with a bleed resistor across the main rail before touching anything, bleed `~2 minutes`). The PSU fan is typically an `80×80×25 mm` or `120×120×38 mm` `12 V` ball-bearing unit. Replace with matching spec. This is Tier 3 because PSU internals are non-trivially dangerous; if you have not worked inside a server PSU before, stop and ship the PSU to D-Central.

Refresh thermal paste on the MM nearest the noisy fan. If Step 7's `PVT_T` cross-check showed a `>10 °C` spread on the noisy-fan side, paste has likely taken thermal damage from repeated overshoots. Remove the MM, clean existing paste with `99%` IPA + lint-free wipes, reapply Arctic `MX-6` or Thermal Grizzly Kryonaut — thin uniform layer, don't glob. Replace any thermal pads on the voltage-regulation ICs that look dried out or compressed. This work is cheap insurance once the fan is fixed.

Check MM-side voltage-regulation caps. Visual inspect electrolytics and MLCCs around the MM power stage on the hot-side board. Bulged tops, vented cans, brown crusty residue, or cracked MLCCs near the PMIC = replace with matching-spec parts. This is soldering iron + hot-air work. Use matching footprint and voltage-rating. Do not guess on cap values — photograph the original or pull the value from Zeus Mining's A1246 hashboard repair documentation.

Roll firmware back one version if the buzz started after a flash. Canaan's signed-firmware model resists rollback, but `avalonminer.org/firmware-document/` publishes multiple build numbers. If `Fan1` / `Fan2` RPM began oscillating after a firmware update and hardware is otherwise clean, the PWM control curve in the new build may be the issue. Flash one version back; observe 24 hours. Use a wired Ethernet connection during flash — Wi-Fi drops mid-update can brick the controller on Canaan's signature check.

Stop DIY and book D-Central ASIC Repair when: (a) Step 10's fan-swap proves the socket is dead and PWM driver IC replacement is required, (b) Step 15 shows multiple bulged caps or visible PCB heat damage near the MM power stage, (c) the PSU internal fan shows `PS[0] = 2048` and you're not comfortable opening a `3250 W` server PSU, (d) you see any burnt-component odour, arc marks, or `PS[0]` bits `128`/`256`/`512`/`1024` set consistently — those are PSU channel-level faults that require bench parts. [Book an ASIC Repair slot →](https://d-central.tech/services/asic-repair/) — our bench runs the A3206 test fixture with programmable AUC3 load, does chip-level reflow with calibrated hot air, and burns every repaired A1246 in at nameplate `90 TH/s` for 24 hours before return shipping. Turnaround 5-10 business days, Canada / US / international.