Avalon 1346 – PSU Voltage Sag Under Load

Hard power-cycle at the PDU for 5 minutes. Full power-off so the PSU's primary and secondary bulk caps fully discharge and the hashboard PMICs cleanly reinitialize. A surprising percentage of "rail sag" tickets are one-time PMIC latches that clear with a real reboot — not a soft restart from the dashboard. If the symptom does not return inside 48 hours of power-up, you may have caught a one-shot transient; monitor before declaring it fixed.

Pull the API `stats` on port 4028 and record a baseline. `echo -n '{"command":"stats"}' | nc <miner-ip> 4028`. Note `PS[0]`, `ECHU`, `PVT_V` per board, `MW0/MW1/MW2`, and timestamp. Without an API baseline every later step is a guess. The A1346 dashboard does not surface `PS[0]` decode or `PVT_V` scatter in actionable form; the API does.

Multimeter on DC, probe at the hashboard end of each PSU-to-hashboard cable while the miner is hashing flat-out. Target `≥ 11.7 V` minimum, no excursion below `11.4 V`. Below those thresholds and you've confirmed rail sag — your subsequent steps narrow which segment is responsible. Use min/max recording mode if your DMM has it; the trough is what matters, not the rolling average.

Confirm the IEC C19 power cable is rated `≥ 16 A` (stamped on the jacket) and seat both ends fully. Aftermarket 13 A or 15 A IEC cords undersize the input, drop voltage as I²R loss in the cord, and look exactly like a tired PSU. Replace any under-rated cord with the OEM Canaan `≥ 16 A` cable or an equivalent.

Multimeter on AC at the wall outlet under full miner load. Target `≥ 220 V` sustained. A 240 V split-phase that sags to `215 V` evening peak silently caps A1346 chip frequency before any rail trip — the firmware chip-frequency table is voltage-aware. If the wall is sagging, fix the wall before suspecting any miner hardware. The miner is not the problem; the circuit is.

IR thermometer on the IEC C19 cord, the PSU casing, and each PSU-to-hashboard cable jacket while hashing. Anything past `50 °C` is dropping voltage as heat — that voltage is being stolen from the rail. Hot cable = sagging cable, every time. The OEM Canaan cables run cool when correctly sized; aftermarket low-quality cables run hot.

Re-seat every PSU-to-hashboard cable with the system powered off. Pull each cable, visually inspect pins for oxidation (green), blackening, or backed-out female pins. Clean contacts with 99% IPA on a lint-free wipe. Reseat firmly until the latch clicks. Connector oxidation in a humid environment (coastal BC, Maritimes, unconditioned basement) climbs contact resistance over 12-18 months and produces sag identical to a tired PSU — without the PSU ever being the problem. If a connector face is visibly damaged, replace the cable.

Swap each PSU-to-hashboard cable end at the PSU side to a different output port on the same `PSU3400`. Power up, hash for 30 minutes, re-pull DMM under load on the suspect rail. If the sag follows the port = bad PSU output channel; if the sag stays with the cable = bad cable. This 5-minute test isolates whether the failure is in the cable or in the PSU's per-channel output stage and saves you from buying a wrong replacement part.

Move the rig to a dedicated 240 V split-phase circuit if currently on 120 V or on a shared 240 V leg. An A1346 on 120 V draws `~28 A` continuous (at the breaker, before any spike) — over a 30 A circuit derate and a code violation in most North American jurisdictions. The same miner on a clean 240 V draws `~14 A`, halves the I²R loss in every conductor and connector, and gives the PSU `>30%` more headroom against any AC-side sag. This is not a luxury; it is table stakes for serious A1346 operation.

Plug-in voltage logger at the wall for 24-72 hours. Capture min/max AC. Healthy 240 V circuit holds above `225 V`. If the line drops below `220 V` correlated with miner symptoms, the circuit is undersized — call an electrician for a dedicated 30 A or 50 A run, or move to a different breaker. The miner is fine; the wall is not. This single $40 logger has resolved more "mystery PSU failures" in our queue than any other diagnostic tool.

Tune cgminer to a lower-power firmware mode per the Canaan 1346/1366 manual (`24.6 J/T` low mode at `~3000 W` wall, vs `26.9 J/T` high mode at `3530 W`). Run for 30 minutes and re-pull rail readings. If the sag clears at lower power, the failure is load-driven (PSU, cable, or AC supply at marginal headroom) — not a static rail problem. Use the lower power mode as a temporary mitigation while you fix the underlying cause.

Verify the `PSU3400` model number on the PSU label. If a previous owner or repair has substituted an A1166- or A1246-class PSU (lower nameplate) into the A1346 chassis, the substitute is not rated for `~3300 W` continuous and brownouts on every spike. Replace with a real `PSU3400` (or, in some late-batch A1346 SKUs, the `PSU3700` cleared by Canaan as compatible). See Avalon — PSU Incompatible Model for the model-matching logic.

Scope the rail at the hashboard connector under full load. AC-coupled probe, `100 mV/div`, `1 ms/div`. Capture peak-to-peak ripple and any deep dips. Healthy: `< 120 mV pp`. Degraded: `> 300 mV pp`. Catastrophic: visible spikes below `11.0 V`. Repeat at the PSU casing. If the casing is rippling badly, the `PSU3400` output filter caps are dry. If the casing is clean and only the connector ripples, it's all cable/connector. The cheapest credible scope today is an FNIRSI 1014D at `~CAD $200`.

Pop the `PSU3400` cover (after a 10-minute unpowered cooldown plus deliberate bulk-cap discharge with a `10 kΩ 5 W` resistor — 30 seconds, then DMM-verify `< 10 V` before any contact). ESR meter on every electrolytic — primary-side mains-rail caps, secondary-side output caps, bias / standby caps. Healthy: `< 30 mΩ`. Marginal: `30-100 mΩ`. Failed: `> 100 mΩ` or any visible bulge / weeping. Replace failing caps with same-voltage, same-or-larger capacitance, low-ESR series (Panasonic FR/FM/FC, Nichicon PW, United Chemi-Con KZE/KZH). On a `PSU3400` 4+ years in continuous duty expect 4-8 caps near end-of-life.

Replace `PSU3400` output cabling with verified `≥ 16 AWG` PCIe sets — one cable per hashboard, no daisy chains, no Y-splitters. Aftermarket low-cost cable kits often ship 18 AWG marketed-as-16 AWG; verify by stripping a cm and measuring conductor diameter (16 AWG = `~1.29 mm`, 14 AWG = `~1.63 mm`). For an A1346 pulling `~1100 W` per hashboard, 14 AWG is the right answer, not the optional answer. Set length should be the shortest workable — every metre of `16 AWG` adds `~5 mΩ` per conductor.

Replace input-filter capacitors on a hashboard if the rail at the connector is clean on the scope but `PVT_V` telemetry still shows scatter and brownout symptoms persist. The bulk SMD electrolytics nearest the 6-pin power input on the hashboard have aged out independently of the PSU. ESR-test in-circuit (rough but workable) or pull and measure off-board. Polymer replacements (Panasonic POSCAP, Sanyo OS-CON) are dimensionally compatible and dramatically improve transient response — recommended over electrolytic if your hot-air station can place them cleanly. SMD rework, not a reflow job.

As a workshop diagnostic — not a permanent fix — solder a `2200 μF / 16 V` low-ESR electrolytic across the 12 V / GND at the hashboard connector terminal on its own short leads. This adds local reservoir capacitance for spike currents and will mask a marginal cable's voltage drop temporarily. Use this to *confirm* the diagnosis (if symptoms clear with the bandage cap, you've narrowed it to insufficient transient capacitance somewhere upstream); then upgrade the cable / cap bank properly. Do not leave the bandage in place long-term.

Reduce A1346 firmware power mode permanently if your environment cannot deliver clean `220 V AC` and a fresh PSU is not in budget. Canaan documents at least three power modes in the 1346/1366 manual; the lowest-J/T mode pulls `~3000 W` wall and gives the PSU and the AC circuit substantial headroom against sag. You'll lose `~5-7%` of nameplate hashrate but you'll keep the rig hashing reliably at peak hours instead of brownouting into the pool's stale-share bucket every evening. Document the mode choice in your operations log.

Stop DIY and ship to bench when any of these are true: bulk-cap replacement cleared ESR but ripple under load remains excessive (output inductor / controller IC / feedback-loop fault); a known-good `PSU3400` swap still leaves rails sagging (multi-board input-filter degradation or chassis wiring fault); visible scorching, burnt-component smell, or capacitor venting on the PSU primary side (fire / shock risk — do not power on); the PSU trips its own thermal protection within minutes under load. Book a D-Central ASIC Repair slot at https://d-central.tech/services/asic-repair/. Bench process: programmable DC load + cap-bank refresh kit + `PSU3400` factory test profile, full primary-side cap audit, primary-side fan replacement if duty is elevated, output-filter inductor sweep, 24-hour full-load burn-in at nameplate to confirm before ship-back. Canadian turnaround 5-10 business days. Include API `stats` dump, scope captures, and AC voltage log in the shipment note.