Antminer L7 – Fan Speed Error

Hard power-cycle the miner at the wall or breaker for 60 seconds. Not a soft UI reboot — a full wall disconnect that drains residual rail charge and forces the control board's fan-controller state machine to cold-start. Clears wedged states from a prior firmware update, voltage brown-out, or a transient dust-ingestion spin-stall. If all four fans spin up above `5000 rpm` and hold for 10 minutes, the fault was transient. Log the event, check again tomorrow, and order a replacement fan set if the pattern returns — transient fan faults on the L7 almost always recur within 30 days.

Capture the Kernel Log from the L7 web UI (Miner Status → Kernel Log) or SSH in and grep `/var/log/messages` for `fan`. Note which fan number (`0-3`) is failing, what RPM it reports at the moment of failure, and the timing pattern — cold boot only, after a specific runtime, or mid-mining. Screenshot the log. Each timing pattern points at a different root cause: cold-boot-only = bearing cold-stall, mid-mining-only = thermal expansion in a worn bearing, intermittent = FG wire or connector. Skipping this step costs real bench time later.

Verify ambient intake temperature and airflow. Use an IR thermometer at the L7's front grille — ambient above `30 °C` accelerates bearing failure; ambient above `35 °C` can push a marginal fan past threshold mid-mining. Clear filter clogs, vacuum intake, ensure `~15 cm` clearance in front of the miner, and wait for ambient to drop. Reboot and monitor. A 5-minute check that has resolved genuine fan-speed events on hot summer days before any parts are replaced.

Power off at the breaker. Wait 60 seconds for capacitor discharge. Phillips #2 to remove the L7 top cover. Work on an ESD mat with a grounded wrist strap. Visually inspect each fan blade for dust mat, hair, insulation, spider webs, or any foreign object wound around the hub. Clear anything obvious with compressed air (short bursts, hold the fan still to avoid spinning past rated RPM) or a shop-vac at low suction. Take photos before disconnecting anything — reassembly at midnight with tired eyes goes better with a reference.

Spin each fan by hand with the power off. A healthy L7 fan spins 3-5 rotations on a flick and coasts smoothly. You are feeling for grit, stiffness, rough rotation, a bearing that resists past a certain angle, or any audible grind. Rough-spinning fan = bearing wear and the fan is on borrowed time even if it's not the one currently throwing the fault. Note every marginal fan and plan to replace the full set. Fans that stop within one rotation are already dead — don't bother retesting them.

Inspect every 4-pin fan connector at both ends. At the control-board end, check that each header is fully seated and the pins are not oxidized or green. Gently tug each connector — movement before release tension means a partially seated connector. Unseat each, inspect the pin rows for corrosion, reseat with a firm click. Do the fan-side connector on each fan too. Use a shop-vac at low suction to remove dust without pulling on wiring. Reboot and watch the log — connector seating alone resolves a surprising number of L7 fan faults.

Swap the failing fan into a known-good header and a known-good fan into the failing header. Cross-connect to isolate fan-vs-header. Reboot. If `ERR_FAN_SPEED` moves to the newly-populated header (following the suspect fan), the fan is dead — order a replacement and skip to step 9. If the fault stays on the original header (good fan also fails there), the control board's fan PWM driver or FG input stage is the problem — escalate to step 13. This single swap test saves more wasted parts than any other step in the sequence.

If the blade is clearly spinning at full speed but the dashboard reads `0 rpm`, test the FG tachometer wire. With a multimeter set to frequency (Hz) or a cheap oscilloscope, probe the FG wire (usually yellow or green on the fan pigtail) while the fan runs. A healthy 4-pin fan produces `2` pulses per revolution — `6000 rpm` = `200 Hz` on the FG line. No pulse with a spinning blade = broken FG wire or dead Hall sensor inside the fan hub. This is the failure mode Bitmain's Tier 1 troubleshooting misses — it will trick you into replacing parts you don't need otherwise.

Source and install a compatible replacement fan. Spec: `120 × 120 × 38 mm`, `12 V DC`, `4-pin PWM`, `~6000 rpm` minimum, `200+ CFM`, high-static-pressure, dual-ball bearing. Stock part: `KZ-12038HA2-P001` or equivalent. Acceptable alternatives: Delta `FFB1212EH`-class, San Ace `9GV1212P1J03`-class, Sunon high-static-pressure axials. Verify the connector pinout (aftermarket fans sometimes swap PWM and FG positions), match airflow direction (intake fans blow into the chassis, exhaust fans blow out), and tighten mounting screws without over-torquing the plastic frame. Reboot and monitor for 24 hours.

If only one fan has failed but others show rough rotation or bearing noise on manual spin, replace the full 4-fan set. Running a fresh fan beside three tired stock fans creates acoustic beat patterns and puts more aerodynamic load on the new fan (because worn fans move less air), which shortens its life too. A matched replacement set also simplifies future diagnostics — all four fans age together. Spend the extra CAD now, save the mid-block `ERR_FAN_SPEED` event later. L7s in dusty garages need this treatment every 18-24 months regardless.

Cross-flash DCENT_OS — D-Central's open-source Antminer firmware — for per-fan RPM visibility, configurable thresholds, and better fault logging than stock Bitmain firmware provides. On stock firmware the fan threshold is hard-coded; on DCENT_OS you set it deliberately per fan, log per-fan RPM history, and alert before a fan drops below threshold (instead of after). Source at `https://github.com/DCentralTech/DCENT_OS`, download at `https://d-central.tech/dcent-os/`. Braiins OS+ / LuxOS / Vnish are the main alternatives — all give better fan telemetry than Bitmain stock.

Temporary FG-spoof bypass for emergency limp-home operation only. If your L7 has one fan whose blade works perfectly but whose FG tach wire is broken, a `1-2 kΩ` pull-up resistor between the FG pin and the control board's `+3.3 V` rail will inject a faked RPM signal that satisfies firmware. This lets you keep hashing for 48 hours while the replacement fan ships. Only use this with a physically-working fan — never to bypass a dead blade. Applying this trick to a stalled fan will cook your `BM1496` chips within minutes. Remove the resistor the moment the new fan is installed.

Measure the control board's `+12 V` rail at a fan header under load with a multimeter. Expected: `11.8-12.4 V` steady, no dips below `11.4 V` during fan startup inrush. If the rail sags, swap to a known-good L7-compatible APW12-class PSU and retest. Rail stable with new PSU = original PSU was sagging under fan inrush, replace PSU. Rail still sags with new PSU = control-board `+12 V` regulator is failing, escalate to step 15. Document the voltage at idle and under load — photos of the multimeter reading help with warranty or D-Central bench diagnosis.

If you are confident with SMD rework and have the right station (hot-air rework, `310-330 °C`, `N` nozzle, flux, kapton tape), component-level repair on the L7 control board is in reach. Common targets: the PWM driver transistor for the failed header, pull-up resistor on the FG line, or the fan-management MCU. Pull the control board, inspect with a USB microscope at 50x minimum. Reflow suspect components one at a time, retest between each. ESD-safe bench, grounding strap, and flux-residue cleanup with 99% isopropyl are non-negotiable. This is Tier 3 work — walk away if you don't have the tools or experience.

Control-board-level faults beyond your comfort zone ship to D-Central for ASIC Repair (`https://d-central.tech/services/asic-repair/`). D-Central's bench keeps stock L7 fans, matched replacement control boards, and PWM driver ICs on hand. Turnaround is typically `5-10` business days from receipt with component-level diagnosis before any parts are replaced — no return-and-swap-the-whole-board shortcuts that send you a different miner's problems. Include a note with the kernel log excerpt and the diagnostic steps you've already completed. Clear communication shortens turnaround.