Avalon 1466 – PSU Voltage Sag

Hard power-cycle at the PDU for 5 minutes — not a soft reboot from the UI. Full power-off lets the PSU3400-01 Plus primary-side caps fully discharge and the AUC / MM controllers re-initialize cleanly. Clears any wedged latch state from a prior BOOTBY[0x21] reboot, drops latched PS[0] status from a transient sag event, and forces the firmware out of any silently-degraded state it might have entered after a series of sag events. If the miner returns clean with PS[0] = 0 and full MW arrays, you have baseline; if it cliffs back to a faulted state immediately, the sag is happening every minute and you'll catch it on the scope.

Pull the CGMiner API stats and version on port 4028 and save the output as your baseline. `echo -n '{"command":"stats"}' | nc <miner-ip> 4028`. Record PS[0], MW0/MW1/MW2, BOOTBY history, firmware version. Compare against this after every subsequent fix step. Without an API baseline, every later change becomes a guess at whether you made things better. The A1466 dashboard alone does not expose these fields in actionable form — the raw API does.

Measure AC at the wall receptacle with a true-RMS multimeter at idle, before the miner is plugged in. Target 230-245 V on residential 240 V split-phase, 205-215 V on commercial 208 V three-phase. If idle voltage is low, the sag-under-load voltage will be worse — you have an electrical-supply problem before you have a miner problem.

Plug the A1466 in, let it ramp to full nameplate, and re-measure AC at the receptacle while it's hashing. Compare delta. A 5 V drop from idle to full-load is normal; 10-15 V drop is marginal; greater than 15 V is unsupportable for an A1466 long-term. The A1466 does not accept 110 V at all (Canaan spec is 220-277 V AC only) — a harder floor than A1246/A1346, which accept down to 200 V.

Confirm the cable from receptacle to PSU is rated and intact. Use the manufacturer-supplied IEC C19 cable; if you've substituted a household IEC C19 jumper, swap it back to a properly-rated cable (Zeus Mining's PSU3500-01 reference shows the spec: 3 × 3.31 mm², 150 cm, IEC C19, rated for the load). Ditch any extension cord under 12 AWG; if you must use an extension, it should be 10 AWG or heavier and as short as physically possible.

Confirm the miner is on its own dedicated 240 V circuit. If it shares a panel circuit with an A/C compressor, electric water heater, electric kettle, or a second miner, those loads' inrush sags the shared feeder every time they cycle — and the A1466 throws PS[0] != 0 events you can never replicate while watching the dashboard. The fix is electrical, not firmware: add a circuit. Single most-common Tier 1 finding in our queue from Canadian residential operators.

Deploy a 24-hour voltage logger on the same receptacle (HOBO MX1102, Extech 380803, or any logger with 1-second AC RMS resolution for 24+ hours). Run it for at least one full cycle of the suspected pattern: 24 h for daily neighbourhood-peak suspicion, 7 days for weekly utility-balancing patterns. Cross-reference the voltage trace against BOOTBY[0x21] events in the API. Aligned events convict the AC feed.

Two-channel oscilloscope on the PSU output rail at the board-side connector. Probe board 1 first — typically the longest cable run and worst-case for sag. DC-coupled, 2 V/div, 2 ms/div. Capture rail behaviour at idle, under steady-state load, and during a known transient (start-up, fan ramp, frequency step). Healthy A1466 rail: 14.7-14.9 V idle, 14.0-14.4 V steady-state, transient dip no deeper than ~13.8 V on a 3-5 ms window, full recovery within 10-15 ms. Sustained dip below 13.5 V for more than 10 ms is in A3198 PMIC UVLO territory and convicts the PSU output.

Re-seat every PSU-to-board cable with the system powered off at the PDU. Pull each connector, visually inspect pins for oxidation, blackening, pitting, or bent pins. Clean contacts with 99% isopropyl alcohol on a lint-free wipe. Reseat firmly until the connector clicks. Oxidized connector resistance adds series voltage drop under load that looks identical to a PSU sag — and on humid coastal or basement deployments, oxidation can develop in 12-18 months.

Replace the AC C19 input cable with a known-good Canaan-rated cable (or equivalent: 12 AWG minimum conductor, IEC C19, length matched to the run). The rated cable is not optional on an A1466 — household-grade IEC C19 jumpers cannot sustain the steady-state current draw without resistive heating that adds further sag.

Verify the PDU / power strip (if any) is rated for at least 30 A continuous on the relevant phase. Residential 15 A smart power strips, surge bars, and KVM power management hardware are unsuitable for an A1466. The PDU's internal MOSFETs, fuses, and contactors all add resistance — cumulative effect at A1466 current levels is measurable sag. Land the miner directly on a hard-wired receptacle whenever possible.

Swap in a known-good PSU3400-01 Plus (from a healthy A1466 or D-Central's stock). Repeat the scope capture in Step 8 with the new PSU. If rail behaviour normalizes — flat 14 V rail, clean transient recovery, no UVLO-region dips — the original PSU is the problem. Cap-aged PSUs are common at the 18-30 month mark in residential deployments, 12-18 months in hot-room commercial. Do not substitute a PSU3300-A02, APW3++, APW9, or earlier-generation PSU into the A1466 — they cannot sustain 3230 W+ and will fail in service.

Inspect the PSU3400-01 Plus casing for visible damage: bulging side panels (primary-side cap pressure), discolouration on the exhaust grille (heat damage), burnt-electrolytic smell (cap venting), audible coil whine that tracks load (regulation loop instability from cap drift). Any of those symptoms means the PSU is on its way out even if it's still passing scope tests today — schedule replacement before it fails outright.

Open the suspect PSU3400-01 Plus and inspect the primary-side electrolytic capacitor stack. The bulk caps that smooth the ~380 V PFC bus are the most-common failure point at the 18-30 month mark. Bulging tops, vented vents, or visible electrolyte under the cap base are replace-on-sight. Match capacitance, voltage, ripple-current, and temperature rating (most stock caps are 400 V 105 °C Japanese-grade). High-voltage work — the 380 V PFC bus stays charged for minutes after AC removal. Discharge through a 10 kΩ 5 W resistor before touching anything inside. If you are not comfortable with primary-side switching supply rework, stop here and ship the PSU to D-Central or replace it.

Inspect and replace board-side input filter capacitors on any A1466 hashboard showing local sag-tolerance issues even with a confirmed-good PSU. The A1466 board's input rail is filtered by a stack of bulk electrolytics and ceramic MLCCs at the PSU connector. Bulging electrolytics, cracked MLCCs, or discolouration near the rail is replace-on-sight. SMD rework: iron + hot-air, correct-value capacitor stock (16 V rated, low-ESR, sized to original spec). Continuous 60-75 °C board operation drifts these caps in 18-30 months. This is component-level work, not reflow — if you're not comfortable with SMD desolder/resolder, ship to D-Central.

Refresh thermal paste on the PSU's primary-side heatsink and any board-side DC-DC heatsinks. Dried paste at the PSU's primary FETs raises FET junction temperature, raises ambient inside the PSU enclosure, accelerates cap drying, and tightens the regulation envelope. Arctic MX-6 or Thermal Grizzly Kryonaut, uniform thin layer. Preventive maintenance more than repair, but on a PSU you suspect of cap aging it buys regulation headroom for the time it takes to plan a full recap or replacement.

Roll firmware to a build confirmed by the BitcoinTalk Avalon A14 thread as good for your specific A1466 hardware revision (control board sticker, cross-referenced against avalonminer.org/firmware-document/). Some firmware builds ship with PSU-control regression bugs that misregulate rail voltage under load. Canaan's signed bootloader blocks downgrade on most A1466 batches — if the current build is a regression and rollback is blocked, document the build version, flag the unit for D-Central bench recovery, and proceed to Step 19. Do not attempt unsigned firmware flashes; bricking via signature mismatch is a ship-to-bench event.

If PVT_V scatter on the API shows board-level voltage drift greater than 30 mV across chips on the suspect board, the local DC-DC regulators or PMIC are aging, increasing the board's UVLO sensitivity. Replace the worst-drifted PMIC if you can identify it on schematic; otherwise replace the local DC-DC inductor and capacitor stack as a unit. SMD rework job. Realistically this is a bench operation — escalate to Tier 4.

Stop DIY and ship to bench when known-good PSU swap did not clear the fault, PVT_V scatter exceeds 30 mV on the suspect board, capacitor bulging or burnt-component smell is present, signed firmware blocks a needed rollback, the miner took a physical event (lightning, arc, drop), or you have chased the same sag for more than 8 hours. Book a D-Central ASIC Repair slot at https://d-central.tech/services/asic-repair/. Bench process: programmable AC source 277 V to 200 V sweep, full primary-side recap, board-side input filter audit, DC load test, 24-hour burn-in at 150 TH/s with API logging. Canadian turnaround 5-10 business days; include API stats, BOOTBY history, and voltage logger trace in the shipment note.