ASIC Miner – Coil Whine from Hashboard

Phone-mic the miner in three states and save dated audio files. Record 30 seconds at idle, 30 at stock-profile hash, 30 at your tuned profile. Label coil-whine-baseline-<serial>-<YYYYMMDD>.mp4 and save to durable cloud storage. Without a baseline you can't answer the only question that matters — 'has the noise changed?' — and you'll guess wrong. Five minutes of work now avoids months of ambiguity later.

Revert to stock profile (no OC, no UV, no autotune) via the miner's web UI and observe for 15 minutes. On Antminer stock firmware: Miner Config → Miner Type → stock. On Whatsminer: Config → Miner Type → stock. If the whine is dramatically quieter or gone, the regulator was being pushed by your tuning into a noisy operating point — cosmetic in the overwhelming majority of cases. The tone itself harms nothing; feel free to keep mining tuned.

Check that the miner is mechanically decoupled from resonant surfaces. A hollow wood table, plywood shelf, or thin sheet-metal rack amplifies board-level vibration like a speaker cone. Place the miner on a 12 mm rubber mat or dense closed-cell foam. If the perceived loudness drops significantly, the root cause was chassis-surface resonance rather than the board itself, and no further diagnostic is required.

Verify ambient temperature at the miner intake is ≤ 30 °C using an IR thermometer held at the front grille, mid-height — not room-centre, not the hallway. Hotter ambient makes the DC-DC stage work harder and raises the coil-whine baseline by 3-6 dB even with no fault present. Address intake temperature before you escalate acoustic concerns; it's a zero-cost diagnostic with a high false-positive rate if skipped.

Check firmware changelogs and roll forward or back one version, observing 30 minutes per revision. Regulator-loop tuning, PWM frequency choice, and load-transient compensation change between firmware releases — Bitmain stock, Whatsminer stock, DCENT_OS, Braiins OS+, LuxOS, and Vnish have all shipped updates that measurably shifted coil-whine baseline. Compare each revision against your Step 1 baseline recording.

Swap hashboards between slots to isolate board vs slot. Power off at the breaker. Label slots 0/1/2 with painter's tape. Move the whiniest board to a known-silent slot. Power up. If the whine follows the board, the board has the issue; if the whine stays in the slot, the control-board / cable / chassis-mount path is the issue. Expect fault-follows-board >90% of coil-whine cases in D-Central's repair queue.

Re-seat every hashboard data and power connector. Power off at the breaker. Disconnect ribbons and power cables at both ends, inspect contacts for blackening, corrosion, or bent pins, reconnect firmly with an audible click. An intermittent power-rail connection causes the buck-regulator loop to hunt, and that hunting manifests audibly as coil whine. Re-seating alone resolves a surprising percentage of mystery whine tickets.

Measure PSU output voltage at the board connector under full hashing load. Multimeter on DC, probe at the board-side connector. Expect ≥ 13.8 V sustained on standard S19 / S19 Pro APW12 rigs; ≥ 13.8 V on S21-series APW171215a; ≥ 12.0 V on L7-class. Low input voltage forces higher buck current, stressing inductors and MOSFETs and amplifying coil whine. Swap to a known-good PSU before chasing a phantom fault on the hashboard.

Tune the profile back one step if the whine appeared after an OC or autotune run. Drop 100 MHz off the clock, or back off -10 mV on an undervolt profile, and observe 15 minutes. On per-chip-visible firmware (DCENT_OS preferred on Antminer, or Braiins OS+ / LuxOS / Vnish), pull the specific offending board back more aggressively than its peers — it's the weakest link in the silicon lottery and it is telling you so.

Replace the PSU-to-board cable if thin-gauge, long, or aged. Resistive voltage drop under load forces the buck regulator to work harder, which amplifies whine. D-Central's bench standard is AWG 10 or heavier for APW12-class rigs. A $20 cable replacement often resolves complaints that looked like a $400 board problem, especially on longer cable runs in basement-to-garage residential setups.

Flash DCENT_OS (D-Central's own open-source Antminer firmware — the Mining Hackers' choice: full per-chip HW% visibility, per-domain tuning, autotuning, stratum v2, open-source on GitHub). Alternatives: Braiins OS+, LuxOS, or Vnish 1.2.x+. All give per-chip HW% and per-board tuning — the single most important diagnostic upgrade beyond the basics. On Whatsminer use MinerTool + the Whatsminer-Tool CLI; on Bitaxe, AxeOS 2.x exposes per-domain telemetry natively. Stabilise 20 minutes before measurement.

Re-apply thermal interface material on the implicated buck domain. Dried or missing thermal pad material between inductors / MOSFETs and the heatsink bar pushes ambient up by 10-20 °C, accelerating electrolytic drift and worsening regulator-loop stability. Use Arctic MX-6 on ASICs and Laird tflex HR400 (or equivalent) 2-3 mm thermal pad on inductor / MOSFET clusters. Uniform layer, no gaps, don't glop.

Replace suspect output capacitors on the implicated buck domain. Drying electrolytics near the buck output alter loop compensation, sloppify regulator response, and amplify coil whine until output ripple exceeds ASIC core tolerance and HW% spikes. Identify output caps using public repair-guide schematics (Zeus Mining's Bitmain series is the most public reference). Replace with 105 °C low-ESR polymer or electrolytic equivalents. Verify with a scope (see Step 7 above — ripple <50 mV pk-pk on the core rail).

Reflow or replace the implicated inductor if stethoscope-localised and thermally hot. Remove, inspect for cracked core, lifted pads, degraded solder fillets. A hot-air rework cycle at 310-330 °C for 30 seconds often re-seats a fatigued solder joint and the whine stops permanently. If the inductor itself is damaged, desolder and replace with the manufacturer's equivalent — Bitmain S19-class uses 6R8 ferrite power inductors (Zeus APW12 / L7 hashboard guides as public reference).

Flash the previous known-good firmware if the whine started immediately post-update. Roll one version back, observe 30 minutes, compare to your Step 1 baseline recording. Firmware-induced coil whine is real and documented across multiple Bitmain and Whatsminer releases. If rolling back cures it, file a bug with the firmware project, stay on the previous version until a fix lands, and document the exact build string in your maintenance log.

Stop DIY and book a D-Central ASIC Repair slot when (a) whine is accompanied by burning smell, scorching, or cap bulging; (b) per-chip HW% isolates dead chip positions downstream of the whiny domain; (c) the offending inductor has lifted pads or cracked silver beneath it; (d) you've reflowed once and whine returned within 30 days; or (e) scope shows sub-harmonic instability or spiky ripple on the core rail. These are bench-repair conditions and further DIY risks destroying adjacent components.

D-Central bench process: full board-level inspection under 10x microscope, per-inductor acoustic signature capture, thermographic map at full nameplate load, scope analysis of every buck-domain core rail, capacitor ESR audit, PMIC / buck-controller swap where indicated, chip replacement with graded BM-family or Whatsminer-family chips, 24-hour nameplate burn-in post-repair. Audio baseline kept on file and post-repair signature confirmed against a healthy peer board.

Ship hashboards safely: antistatic bag per board, bubble-wrap over the bag, double-box with ≥ 5 cm foam on every side. Include a packing note listing observed symptoms, baseline audio file link (Google Drive / Dropbox), firmware version and profile, ambient temperature at install site, and any DIY steps attempted. Canadian customers: Purolator ground to the server2.d-central.tech address on your ticket. US and international welcomed; shipping quoted at intake.