Avalon 1446 – Fan Speed Error

Hard power cycle at the breaker. Flip the breaker feeding the 1446 (IEC C19 input, ~2800 – 3000 W nominal — a dedicated 240 V circuit is strongly preferred). Wait a full 60 seconds for 12 V rails to discharge. Bring it back. Watch the UI during the first 3 minutes post-boot. Phantom `FAN_FAULT` events from AUC3 I²C timing glitches frequently clear on a cold start and don't return until the next ambient spike. If this clears the fault, log the ambient temperature at failure — you have a marginal thermal environment, not a hardware failure, and the real fix is better airflow or a lower performance bin.

Visual inspection of all four positions. Unscrew the chassis top (four Phillips screws — stock 1446 uses M3 with blue thread-lock; replace any you strip). Shine a phone flashlight across both rotors of each `DF1205012B2FN`-family assembly. Look for dust buildup on blades and around the hub magnet, objects wedged between the rotors (a zip-tie, a missing washer, a dropped nut from a prior service), discolouration indicating heat-kill on a bearing. Dust on the top rotor is the signature A14 family failure mode — the top rotor is upstream in the counter-rotation pair and eats the intake dust load first.

Clean intake and blow out all four assemblies. Shop-vac the intake grille (front of the 1446 chassis). Then hit each fan with dry compressed air (canned or a dry shop compressor — never wet or unfiltered shop air). Hold each rotor with a chopstick or plastic pick so it does not over-spin — over-spinning a dead-powered fan damages the bearing worse than the dust it's being cleared of. Wipe visible hub surfaces with 99% IPA on a lint-free wipe. Roughly 30% of what Canadian home-miner customers ship D-Central as "dead 1446 fan" tickets are actually "heavily contaminated 1446 fan" tickets that clear at this step.

Check ambient and airflow clearance. Canaan's A1346 / A1366 / A1446 manual family gives operating temperature bands of 45 – 55 °C and 55 – 60 °C for different performance modes — those are chip and PCB measurements, not ambient. Keep inlet ambient at or below 35 °C. Nothing within 15 cm of the intake. If the 1446 is in a closet or a small room and exhaust recirculates, the MM enters a thermal protection state that reads almost identically to a fan-speed fault on the dashboard. Move it, duct the exhaust out of the living space, or ventilate the room. For Canadian operators running the 1446 as a space heater (a very good use of 2800 – 3000 W in winter), this is where you balance heat output against intake temperature.

Re-seat all four fan harnesses. Power off. For each assembly, unplug the 4-pin at the MM header and at the fan extension cable. Inspect pins for greenish oxidation or black tarnish. Plug back in firmly. A drop of dielectric grease on re-seat extends the joint by months. Vibration from 24/7 operation works these connectors loose after 6 – 18 months — in the field this single step catches a surprising share of persistent `FAN_FAULT` tickets, especially on units power-cycled several times during shipping or moves between operators.

Measure PSU rail under load. Multimeter on DC. Probe the 12 V rail at the hashboard input while the 1446 is fully hashing (or the closest-to-full-hash state before the fault triggers). Expect sustained 12.0 – 12.3 V. Anything under 11.6 V sustained means the `PSU3300` / `PSU3400` is sagging — fan tach signals ride on rail stability, and a tired PSU produces erratic RPM reads that mimic a dying fan. Swap PSU with a known-good unit to isolate. Confirm input voltage is 200 – 300 V AC (dedicated 240 V strongly preferred — 208 V commercial stacks are marginal, 120 V single-phase is out of spec).

Swap the suspect fan with a known-good unit. Spec: 12 V, 120 × 120 × 50 mm, dual-rotor counter-rotating, 4-pin PWM, high static pressure, rated for 1346 / 1366 / 1446 / 1466 / 1566 chassis — part families include `DF1205012B2FN` (Martech, primary) and the earlier `HA1250H12SB-Z` (cross-ships on some revisions). Zeus Mining and bit2miner stock at ~$8 – $12 USD per unit. Generic single-rotor 120 × 38 PC fans will NOT work — insufficient static pressure through the three-hashboard stack means cascading thermal faults. Secure all four Phillips screws on install.

Test each fan across all four MM headers. Power off. Move one known-good fan through headers 1 → 2 → 3 → 4, powering on briefly at each position to read RPM. If a specific header never reads RPM regardless of which fan is plugged in, the MM tach circuit for that channel is dead (Tier 3 or Tier 4). If a specific fan never reads RPM regardless of header, the assembly is dead (swap). This isolation takes 10 – 15 minutes and eliminates the biggest "is it the board or the fan" ambiguity faster than any other single test.

Verify PSU fan with a `PS[0]` query. Use the `ascset` / `ps` command sequence over port `4028` to pull the PSU status bitmap: `echo '{"command":"ascset","parameter":"0,ps,0"}' | nc <miner-ip> 4028`. Decode bit 11 (`2048` decimal = `0x800`). If set, the failure is inside the sealed PSU, not the chassis. Record the `PS[0]` value and ambient temperature at fault — useful context if you ship the PSU to D-Central. On the 1446, the PSU fan is accessed only by opening a sealed housing. Do not do this unless you have live-rail experience with 200 – 300 V AC bulk caps.

Clean the AUC3 contacts and re-seat. Power off. Unplug the AUC3 from its USB host and from the MM's 2×5 IDC ribbon. Clean both contact rows with 99% IPA on a dental pick (gentle — the plating on AUC3 IDC pins is thin). Re-seat firmly. A tired AUC3 header produces I²C bus errors that surface as phantom `FAN_FAULT` events, especially under high ambient where bus capacitance shifts. Free, 5-minute check that has solved roughly one in twenty of the 1446 fan tickets we've seen inbound.

Flash a known-good Canaan MM firmware. Download signed 1446-family images from `avalonminer.org/firmware-document/` (1346 / 1366 / 1446 MM branch matched to your hardware revision). Flash via AUC3 utility in an idle state. Observe 30 minutes for phantom faults. If the fault clears on a one-version-back build, log the buggy version string. Note: `DCENT_OS` — D-Central's open-source Antminer firmware — does not run on the 1446. Wrong silicon family entirely. `DCENT_OS` coverage for Canaan hardware is on the D-Central roadmap but not shipping today. Do not trust unsigned third-party 1446 MM builds unless you compiled them yourself.

Tune the AUC3 I²C bus timing. If phantom `FAN_FAULT` events persist at high ambient even on a firmware rollback, the AUC3 is sampling the tach on the wrong PWM edge. In `cgminer`, set `--avalon7-aucspeed 400000 --avalon7-aucxdelay 24` (defaults `400000` / `40`). Lower `aucxdelay` tightens the sampling window. Undocumented in Canaan's public material, covered in the ckolivas `cgminer` `ASIC-README` for the Avalon7/8/10/13 family — the 1446 is in this `avalon7-*` command tree because it shares the AUC3 path with every A12 – A15 model.

Replace the MM fan header MOSFET or tach pull-up. If Step 8 identified a dead header, the PWM-side MOSFET or the tach-line pull-up resistor is usually the culprit. Use a 3× loupe or USB microscope to locate the failed part (discolouration, cracked package, blackening at the pad). Desolder with hot air at 300 – 320 °C, replace with same package and spec, verify continuity with a multimeter before re-powering. Expect 30 – 60 minutes of bench work per header and SMD rework experience on QFN / small SOT packages. The Zeus A11 / A12 hash-board repair guide has the closest public reference schematics.

ASIC chip-level awareness (context, not a fan fix). If the `FAN_FAULT` has been running long enough to cook the hashboard, the downstream failure is usually a specific A3200CFA chip in the middle of the longest heat plume — typically mid-board, mid-chain. Zeus's A3200CFA replacement tutorial covers the chip swap; the BIN number on the replacement chip must match the BIN silkscreen on the hashboard or you'll see `MW` / `ECHU` cascade faults on boot. This is a bench-level job. We mention it here so you know what you're buying if you run the 1446 past the fault.

Inspect MM rail regulators. A damaged 12 V → 5 V or 12 V → 3.3 V buck on the MM makes the tach-reading MCU interpret rail noise as RPM fluctuations. Measure at test points. Any rail more than ±5% off nominal (5 V rail: 4.75 – 5.25 V; 3.3 V rail: 3.13 – 3.47 V) means the MM needs component-level repair or replacement. Outside those windows, stop and ship — you're chasing a rail problem, not a fan problem, and the MCU will keep giving you false positives until the rail is clean.

Know when to stop. Ship the 1446 to D-Central when any of the following is true: `PS[0] = 2048` PSU fan failure without live-rail PSU experience; more than one MM fan header dead simultaneously; replacement `DF1205012B2FN`-family fan reports the same fault on the same header; visible MM damage (burnt components, bulging caps, scorched traces); bricked AUC3 or MM after a flash attempt; cascade faults combining `FAN_FAULT` with `ECHU` and rising `PVT_T`. [Book an Avalon repair slot →](https://d-central.tech/services/asic-repair/). Canada-wide, US and international welcomed.

What D-Central does at the bench. Test fixture with programmable 200 – 300 V AC input and per-channel fan-harness isolation for sustained fault reproduction. Logic-analyzer capture of the AUC3 I²C bus for timing bugs. Component-level MM repair (tach MCU, I²C controller, rail regulators, PWM MOSFETs). `PSU3300` / `PSU3400` recap or full replacement on sealed units. Hashboard component-level repair if the fault has been allowed to run long enough to take out A3200CFA chips (BIN-matched replacement). Post-repair 24-hour burn-in at nameplate hashrate before return ship, with `Fan[]`, `PVT_T`, and `GHSavg` logs attached to the ticket for transparency.

Ship safely. Pull the PSU out of the chassis. Pack separately. Pull the three hashboards into anti-static bags. Wrap the chassis in 5 cm of foam on every face. Double-box. Include a plain-text note: observed symptom, intermittent versus constant, current MM firmware version, `PS[0]` value if captured, ambient temperature at fault, contact info, and what you already tried. Every piece of context cuts bench diagnostic time — and diagnostic time is what the repair bill buys. The 1446 is recent enough that customer-shipped context is often the difference between a 5-day and a 10-day turnaround.