Avalon 1246 – High DH% Error Rate

Hard power-cycle the miner at the breaker for a full 60 seconds — power off, capacitors drain, then back on. A soft reboot can carry stale DH% accounting and ECHU bits across the restart; a full cold-start clears cached state and any post-firmware-update glitches. Watch the first 15 minutes of `estats` output after power-up to confirm whether DH% returns or clears. This alone resolves about 8-10% of DH% >2% tickets in D-Central's Canaan queue.

Verify ambient at the intake grille with an IR thermometer — not room-middle. Target ≤ 30 °C at the front face. DH% is a thermal-margin problem at heart; if you are operating at 32+ °C ambient, you have no margin to lose and the flag will keep returning. Move the miner, add airflow, or open a window before pursuing other causes. Canadian summer garages and southern-US deployments are the usual offenders.

Clear the front 30 cm of the intake grille — shelves, curtains, boxes, another miner's exhaust dump. The 1246 is a stacked-fan unit that chokes on restricted intake. Zero-dollar, two-minute fix that resolves a surprising percentage of DH% tickets that appeared after the workshop was rearranged.

Compressor-blow the fin stack and chassis fans. Real air compressor at 80-90 PSI, miner off, blow from the exhaust side back through the intake. Canned air does not move enough volume to clear a 1246 fin stack — you need a real compressor. Spin each fan blade by hand while blowing. Expect a visible dust cloud on the first pass; do a second pass after 30 seconds to confirm no remaining lint.

Reset frequency and voltage to factory defaults from the AvalonMiner web UI: Configure → restore default frequency, restore default voltage, save, reboot. Let the miner run 20 minutes at stock settings. Pull a fresh `{"command":"estats"}` reply and read DH% per chain. If DH% drops below 1.5% per chain at stock, your tune was beyond silicon-lottery limit — rebuild the OC slower in Step 9.

From a laptop on the same LAN, run `echo -n '{"command":"estats"}' | nc <miner-ip> 4028 > dh-baseline.txt` and save the snapshot. Read `DH%`, `PVT_T`, `PVT_V`, `MW0/1/2`, and `ECHU` arrays. Identify which chain (0, 1, or 2) is worst. This baseline is the single most important diagnostic artefact — it is what D-Central's bench will ask for if the fix escalates to us, and it tells you whether the fault is chain-specific or global.

Multimeter on DC, probe at the PSU-to-AUC3 connector while the miner is hashing at full nameplate. Expect ≥ 12.0 V sustained on a healthy 1246. Anything below 11.8 V means the PSU is tired or the circuit is undersized — swap PSU with a known-good unit (Canaan APW-series, 1246-compatible). If DH% clears after PSU swap, you had a sag problem; the original PSU is likely tired and not an immediate liability for a different miner either.

Check line voltage at the panel under load. On 240 V split-phase, expect 235-245 V. On 208 V commercial, expect 202-212 V. On 120 V residential (which the 1246 should not be on, but sometimes is), expect 117-122 V. Low line voltage drives PSU sag, which drives elevated DH%. A $40 voltage logger on the outlet for 48 hours is the cheapest diagnostic in mining.

If reverting to stock cleared DH%, rebuild the tune slowly. From the AvalonMiner UI, raise frequency in 25 MHz increments with 10-minute stability windows between steps. Add core voltage in 25 mV increments only if frequency-only tuning produces DH% >1.5%. Stop at the last step before any chain crosses 2% sustained. That is this miner's silicon-lottery ceiling — it varies per individual unit and per ambient temperature.

Power off at the breaker. Re-seat every hashboard data ribbon and power connector. Visually inspect contacts for blackening, corrosion, oxidation, or bent pins. Wipe contacts with 99% IPA on a lint-free wipe if needed. Reconnect firmly until the latch clicks. A ribbon sitting 0.5 mm proud of seated feeds garbage thermal and work data to the MM3 controller and shows up as DH% noise on that chain.

Label the three hashboard slots 0/1/2 with tape. Move the worst-DH% board to a different slot. Power on, run 20 minutes, pull fresh estats. If high DH% follows the board, the board itself is the problem (paste, chip, or LDO domain). If DH% stays in the original slot, the control path / AUC3 / cable / MM-side is the problem. This 30-minute swap-test saves hours of guessing.

Full thermal paste refresh on the flagged board. Remove the fin stack — Phillips or Torx fasteners, preserve or replace Kapton-backed thermal pads on the PCH and voltage-domain ICs. Clean old paste with 99% IPA and lint-free wipes — no residue on die or heatsink. Apply Arctic MX-6 or Thermal Grizzly Kryonaut in a thin uniform layer (rice-grain blob per die, let mounting pressure spread it). Reassemble with even torque. This restores full thermal margin on the vast majority of 18-24 month old 1246 boards flagging elevated DH%.

Replace crumbled thermal pads on PCH and voltage-domain ICs while the fin stack is off for paste refresh. Crumbled pads transfer heat poorly and raise nearby chip junction temps indirectly through PCB thermal coupling. Match pad thickness with a caliper measurement before sourcing replacements — Canaan does not publish pad specs, but 1.5 mm Arctic TP-3 is the right starting point on most 1246 positions.

Tune AUC3 IIC bus down if Step 11 showed a slot-specific DH% pattern. Edit cgminer args to `--avalon7-aucspeed 200000` (down from 400000 default), leave `--avalon7-aucxdelay 19200`. A slower bus is more tolerant of marginal cables and NTC noise. Watch the log for `CODE_MMCRCFAILED` events — if those disappear and DH% clears, you had an AUC3 bus-margin problem masquerading as a chip-level fault.

Reflow the worst-performing chip on the offending board only if Tier-3 paste refresh was clean and the same chip position is still flagging elevated DH% 2-4 weeks later. Preheat bottom side at ~150 °C, top-side hot air at 310-330 °C for ~30 seconds, natural cooldown, fresh paste on reassembly. The A3206 BGA tolerates one reflow cycle well. A second reflow on the same chip within 90 days rarely holds — replace the chip instead via Tier 4.

MM firmware regression test: note the current MM version from the dashboard, then download one version back from the Canaan firmware portal at canaan.io/support, flash, reboot, observe 30 minutes. Roll one version forward, observe again. Specific MM builds occasionally ship with stratum or work-accounting bugs that elevate apparent DH% — rare but eliminable. If DH% clears on a different version, pin the working version and avoid the bad one.

Stop DIY and book a D-Central ASIC Repair slot when any of these are true: per-chip triangulation isolates the same A3206 position on two different boards; a reflowed chip returns to elevated DH% within 30 days; PMIC or voltage-domain LDO damage is suspected (drifted PVT_V with no thermal correlation, visible heat damage, measurable short); visible capacitor bulging, cracked MLCCs, or burnt-component smell; slot-specific DH% persists after AUC3 swap and bus-speed tuning. Book at d-central.tech/services/asic-repair/ — 5-10 business day turnaround, Canada / US / international.

Ship safely to D-Central. Hashboards in anti-static bags, double-boxed with ≥ 5 cm of foam on every side. Include a physical note inside the box with: full estats baseline from Step 6, MM firmware version, install date, last paste-refresh date, ambient at install location, the Tier 1-3 steps you completed, and which chain(s) flagged. Every minute the bench spends reconstructing fault history adds to the repair bill — a well-documented ship-in saves real dollars.