Antminer L7 – Vibration Noise

Hard power-cycle the miner for 30 seconds at the PDU (not a web-UI restart). Clears any latched fan-PWM state from a transient temp-sensor glitch. If the noise is gone and stays gone through the first hour of full-load operation, it was transient. If it returns, the fault is real — continue to Step 2.

Shop-vac the intake and exhaust grilles with the miner powered off. Dust, lint, pet hair, or insulation fibers drawn into a fan cage cause imbalance that's easily mistaken for bearing failure. A 10-minute clean on a 3-month dust build-up can drop noise 5-8 dB(A) at zero parts cost. Verify no foreign objects inside the fan cages.

Walk every external chassis screw with a Torx T10 driver and tighten anything loose. L7 vibration over 12+ months incrementally backs out top-cover, side-panel, PSU-cage, and mounting-rail fasteners. Loose external screws are responsible for a disproportionate number of 'mystery rattle' complaints — fix this before assuming mechanical failure inside.

Move the miner off any wooden shelf, hollow-core desk, or resonant surface and onto either a dedicated welded steel mining rack, a rubber anti-vibration mat (~CAD $15 washer-pad works fine), or concrete. The miner hasn't changed; the soundboard has. A surprising number of noise complaints resolve at this step with zero mechanical intervention.

Verify intake ambient temperature with an IR thermometer at the front grille (not room-middle). Target ≤ 30 °C, ideally ≤ 25 °C. Ensure ≥ 45 cm clearance in front of the intake and ≥ 30 cm behind the exhaust. A fan at 4,800 RPM in a cool room is quieter than a healthy fan at 6,200 RPM in a hot room.

Log into the L7 web UI, open Miner Status, and record each fan's RPM over 5 minutes at full load. Healthy spread is within ±150 RPM across the four fans. Any fan more than 400 RPM below its siblings is drag-loaded or failing. Cross-reference with the hand-spin ear test from diagnostics — a grinding fan typically reads 3,500-4,800 RPM because bearing drag physically slows it.

Power down at the PDU and wait 2 minutes for PSU caps to bleed. Remove the top cover (8 × Torx T10). Re-seat all 3 hashboard ribbon cables: lift the latch, unplug, inspect for bent pins or blackening, re-plug firmly, re-latch. On the L7 the fan-control loop reads hashboard temp sensors through the ribbon — a marginal ribbon produces PWM hunting that sounds like a failing fan. Re-seat BEFORE swapping any fan.

Replace the confirmed failing 12038 fan. Unplug the 4-pin connector from the control-board header, remove 4 × Phillips #2 screws holding the fan to the chassis, install the replacement (Sunon PF1231512HB, Delta QFR1212GHE, or Bitmain-OEM spare — any dual-ball-bearing 12V DC 12038 rated ≥ 0.60 A and ≥ 220 CFM). Re-seat the 4-pin connector fully. Power on, verify 5,800-6,200 RPM on the dashboard under full load.

Replace the APW12 PSU if diagnostic Step 5 identified buzzing. In-place cap replacement is Tier 3 bench-repair, not Tier 2 DIY. Match the APW12 variant (APW12-1417-A, APW12-15, or APW12-17) to your L7's hashrate bin — the variants differ in voltage rails. Running under-spec PSU on over-spec L7 recreates the sag and buzz you're trying to eliminate. Do NOT keep running a buzzing APW12 in a residence — cap venting is a fire risk.

With the lid off, walk every internal fastener (hashboard mounting brackets, PSU cage, control-board standoffs) with the correct driver — Torx T10 for chassis, Phillips #2 for internal. Tighten, don't crank. Loose internal fasteners are frequently diagnosed as heatsink failures when they are actually just backed-out screws under sustained vibration.

Re-bond a loose or dropped heatsink on a hashboard. Remove the affected heatsink with a plastic spudger (never a metal screwdriver against the PCB). Clean ASIC and heatsink mating surfaces with 99% IPA until mirror-clean. Apply a thin, uniform layer of high-temp structural thermal epoxy (TIM TC-5026 or Arctic Alumina, rated > 200 °C operating). Press heatsink into place, clamp with a non-marring binder clip for the 5-min tack / 24-hour full cure. Post-cure, IR-image under load — chip-to-heatsink delta should be < 8 °C at steady-state 78 °C chip temp.

Replace the whole fan if a blade is chipped — do not attempt field blade repair. A chipped blade creates rotor imbalance that cannot be corrected by trimming or gluing because blade moment-of-inertia is engineered at manufacture. The hour spent attempting a field repair is worth more than three replacement fans. If you find a chipped blade, discard the fan and source a new matching 12038.

If bench-equipped (ESR meter, solder station, hot-air rework, spare 105 °C or 130 °C Japanese-brand electrolytics), recap the APW12. Target caps: primary bulk filter (~220 µF 400 V typical), output rail filters (~3300 µF 25 V typical). Replace with Rubycon, Nichicon, or Panasonic equivalents matching voltage, capacitance, and ESR rating. Budget CAD $20-40 in caps vs $180-320 for a new PSU. If not equipped, ship to D-Central for recap service.

Dampen chassis panel resonance with strategic sound-deadening patches. Apply butyl-rubber-plus-foil automotive mat (Dynamat or equivalent) in 15 × 15 cm patches on the inside of each side panel. Do NOT apply full coverage — full coverage restricts thermal dissipation and can drive chassis temp up. Small patches kill panel-resonance modes while preserving airflow. Can drop perceived noise 3-5 dB(A) with no cooling impact.

For residential L7s, address noise at the room level with a ducted exhaust or a vented acoustic enclosure. In Canadian climates, route exhaust into a basement cold room, attached garage, or the home HVAC return plenum (with a MERV 8 filter to block chip dust). This solves noise AND captures ~3,400 BTU/h of usable heat — the dual-purpose mining angle. None of this fixes a mechanical failure, so complete Tier 1-3 first if a failure is real.

Stop DIY and ship to D-Central Repair when: multiple hashboard heatsinks have delaminated; the APW12 shows cap bulging, discoloration, or electrolyte smell; thermal damage is visible on hashboard pads from running with a failed fan; or you've swapped the fan and the noise / temp anomaly persists within 30 days. You're now in test-fixture territory and need bench tools to isolate the fault. Book a repair slot at https://d-central.tech/services/asic-repair/.

D-Central bench process on an L7 noise repair: IR thermal imaging to map chip-to-heatsink thermal delta per chip position; programmable-load PSU to reproduce symptoms at varied duty cycles; per-chip isolation with Bitmain L7 test binaries; heatsink re-bond with structural thermal epoxy where delamination is found; APW12 recap with graded 105/130 °C Japanese electrolytics; post-repair 24-hour nameplate burn-in with IR monitoring before ship-back.

Ship hashboards in anti-static bags, double-boxed with ≥ 5 cm foam on every side and corner protection. Include PSU if in-scope for repair. Enclose a note describing: noise type (grinding / rattling / buzzing / whining), which fan or hashboard you suspect, current firmware version, whether you've already attempted ribbon re-seat or fan swap, and contact info. This saves our bench 20-40 minutes of re-diagnosis, which directly lowers your invoice.