Avalon 1246 – PSU Not Detected

Kill power at the wall for 60 full seconds. Not a soft reset — fully de-energize the PSU bulk capacitors; count it out. Power back on. A meaningful share of `PSU_ERR` reports on the 1246 are transient supervisor-chip lockouts after a brownout or dirty-power event, and full discharge clears the latch. If the miner comes back clean, log the date and consider adding a UPS or line conditioner — because it will happen again.

Verify the AC cord is fully seated at both ends. Canaan C19/C20 inlets are stiff when new and loosen with age; a 90% seated cord delivers intermittent power, which produces exactly the `PSU_ERR` pattern. Unplug, inspect pins for heat discoloration (a sign of arcing), plug back in firmly until it bottoms out, tug to confirm mechanical retention.

Check your breaker panel and circuit sizing. A 15 A 120 V residential circuit cannot sustain a 1246 at full hash — sustained draw causes brownouts that present as `PSU_ERR`. Move to a dedicated 240 V 20 A circuit (Canadian / US-commercial) or 30 A commercial circuit. This is the single most common mystery-`PSU_ERR` cause on home-mining installs.

Factory-reset the miner via FUNC button — hold for 5 seconds until the LED pattern changes. Clears any stale config that might confuse the PSU polling loop. Not a fix on its own, but rules out software-only state corruption in 30 seconds.

Check ambient environment. Avalon 1246 PSUs are rated for inlet air up to 35 °C; dust buildup or wiring blocking the intake will throttle or lock out the rails and present as `PSU_ERR` that clears on cooldown. Canadian winter operators rarely hit this; summer garage setups should check. Blow out the PSU intake with compressed air (miner unplugged).

Measure AC input at the PSU inlet under load. Multimeter on AC volts, probes on the IEC inlet while the miner is attempting to hash. Expect `220-240 V` (or `100-120 V` for region-matched units). A drop below `200 V` on a 240 V unit means line sag — you will chase `PSU_ERR` forever until the circuit is upgraded. Log the reading; it becomes evidence if you need to escalate to a landlord or electrician.

Re-seat every PSU-side connector. Breaker off, wait one minute. Disconnect the 14-pin signal ribbon, both PSU-to-hashboard output cables, and the AC cord. Inspect each connector under bright light for blackening (arcing), oxidation (humidity damage), or bent pins (prior bad re-seat). Clean with 99% isopropyl on a cotton swab, dry fully, re-seat firmly, listen for the click. This alone clears a significant fraction of `PSU_ERR` tickets.

Test the PSU with control board only — no hashboards. Power off. Disconnect both hashboard output cables at the PSU end, leave the signal ribbon connected. Power on. Healthy behaviour: PSU fan spins at low speed, no clicking, control board boots, API responds. If the PSU enables cleanly here but failed with hashboards attached, the fault is downstream — proceed to Ω-test. If the PSU still refuses to enable, the fault is PSU-side or signal-ribbon — proceed to substitution test.

Ohm-test each hashboard's power input. PSU fully disconnected. Multimeter on resistance, probe V+ to V- on each hashboard in turn. Healthy 1246 hashboard: kΩ range cold. Direct short (< 10 Ω) means a blown input cap or shorted FET on that board — that's the reason the PSU refused. Mark the board, pull it from service, book it for hashboard repair. Run the miner on the remaining boards to confirm the PSU recovers.

Substitute a known-good Canaan/Avalon 3300 W PSU with matching region code and connector set. If the miner comes alive and the API populates `PS[]` with real data, the original PSU is the fault. Decide: replacement unit, or component-level PSU rebuild (bulk-cap replacement is the most common fix and extends life 2-4 more years on this PSU class).

Verify you have the correct PSU for your region and hardware. The single most common DIY mistake on `PSU_ERR` tickets: an Antminer APW3/APW9/APW12 pulled from a parts bin and plugged into a 1246. The pinout does not match, the signal protocol does not match, the control board cannot talk to it. Confirm the label reads Canaan/Avalon spec and region code matches the feed (100-120 V vs 220-240 V).

Inspect the PSU-to-hashboard output cables for crush damage. These run through chassis cutouts and can pinch during reassembly. A partial short in the insulation reads exactly like a dead PSU because modern supplies refuse to enable into a short. Run fingers along the full length, look for discoloration, deformed insulation, or exposed strands. Replace in doubt — they are cheap insurance.

Bench-test the suspect PSU on a DC programmable load. Pull the PSU from the miner; ramp load from 0 A to ~180 A at 12 V. A healthy 3300 W Canaan PSU holds rail voltage ≥ `11.8 V` at full load. Rail sag below `11.5 V` means tired bulk caps on the output stage — you will see `PSU_ERR` intermittently under load even if it passes no-load tests. This is the test that confirms a cap rebuild will actually fix it.

Replace bulk electrolytic capacitors on the PSU output stage. Most common component-level repair on aged Canaan 3300 W units: the large output electrolytics (typically 3300 µF 16 V or 2200 µF 25 V depending on revision) dry out after 3+ years of continuous 80 °C operation. Desolder, test ESR (healthy < 20 mΩ; dried > 100 mΩ), replace with Japanese-brand equivalents — Nichicon, Rubycon, or Panasonic. Turns a $420 replacement into a $30 rebuild.

Inspect the MM3v2 control board PSU interface circuitry. If a known-good PSU fails to register, the control board's interface has failed — common root causes: blown pull-up resistor on PS_ON, fried level shifter on the telemetry line, ESD damage to the supervisor IC. Under magnification, look for burnt or discolored components along the 14-pin signal ribbon trace path. Component-level repair needs the schematic — D-Central has it, Canaan does not publish it.

Explicit note on firmware: DCENT_OS is D-Central's open-source Antminer (Bitmain) firmware and does not run on Avalon hardware. Stick with stock Canaan firmware on the 1246. Community Avalon forks (e.g. EMRYS no-mercy) change mining behaviour, not hardware diagnostics; none fix a PSU polling-path fault. Report bad PSU as a hardware problem — no firmware papers over it.

Flash stock Canaan firmware over AUC3 as a last resort. If all hardware checks out and the API still reports `PSU_ERR`, a corrupted firmware image is possible. Download the latest 1246 firmware from `avalonminer.org/firmware-document/`, flash via AUC3 in recovery mode. Warning: if your signal ribbon is intermittent the flash will fail partway and brick the unit to `FW_ERR` on top of `PSU_ERR`. Fix the cable first, flash second.

Tune AUC3 bus timing if the signal ribbon is physically marginal. CGMiner supports `--avalon7-aucspeed` (default `400000`) and `--avalon7-aucxdelay` (default `19200`) — documented in cgminer's ASIC-README, undocumented by Canaan. Drop aucspeed to `200000` and raise aucxdelay to `38400` for margin on a borderline bus, which sometimes masks as intermittent `PSU_ERR`. Temporary mitigation, not a fix — replace the cable when able.

Know when to stop DIY. If you have swapped the PSU with a known-good unit and the control board still cannot talk to it, the MM3v2's PSU interface is toasted. Component-level bench repair needs a schematic, a stable hot-air rework station, and SMD passives inventory. That is shop work — book a D-Central ASIC Repair slot at https://d-central.tech/services/asic-repair/ and ship the unit.

D-Central bench process on inbound 1246 `PSU_ERR`: known-good PSU swap as step zero to isolate PSU vs. control, full nameplate load-test on the PSU, component-level cap replacement if ESR is out of spec, control-board interface repair from salvaged-grade donor parts (pull-ups, level shifters, supervisor IC) where needed, 24-hour nameplate burn-in pre-return. Turnaround 5-10 business days from receipt; Canada-wide shipping, US and international welcomed.

Ship safely. Strap the PSU so it cannot flop inside the chassis — a dislodged PSU cracks hashboards in transit. Use original Canaan foam if you have it, otherwise double-box with ≥5 cm foam on every side. Include a note with the `PS[]` API signature you saw, firmware version, observed LED pattern, and your contact info. Shaves 30 minutes of diagnostic time off the invoice.