Antminer S9 – PSU Voltage Drop

Hard power-cycle at the breaker for a full 30 seconds, not a soft reboot. Clears any wedged driver state that can mask an underlying voltage issue and resets the PSU's internal undervoltage-lockout latch. Verify the miner is completely dark at the front panel before re-energizing. If the miner comes back up and runs cleanly for 20+ minutes, the issue may have been a transient mains event rather than PSU aging — continue to Step 10 (mains logging) to confirm.

Reset to stock firmware profile. Remove any OC / UV tweaks. S9 stock is ~14 TH/s at ~80 W/TH; S9j ~14 TH/s; S9k ~16 TH/s. Reboot, let the miner run 15 minutes under stock profile, watch for `PSU_ERR` recurrence. If rail sag disappears under stock but returns under any OC, your previous profile was past the silicon-lottery ceiling for the current state of the PSU — the rail margin is too thin for aggressive tuning.

Move the S9 to a dedicated 240 V / 20 A circuit on a NEMA 6-20 or L6-20 receptacle if it isn't already there. This single change solves a large fraction of S9 voltage-drop tickets in D-Central's repair queue. The APW3++ was specced for `200-240 V` operation; `120 V` is tolerated but stresses primary-side components at roughly double the current for the same output power, accelerating cap and MOSFET aging. No other loads on the circuit.

Declutter the cord path between wall and PSU. Remove any C13/C19 adapters, pigtails, multi-outlet splitters, or cheap generator-cable reducers. Factory-style direct connection only — C19 receptacle to C20 inlet on the APW3++, with no intermediate connectors. Cheap adapters introduce voltage loss, poor contact, and hidden fire risk. Inspect the jumper cable for blackened pins, discoloration, or crushed strain relief; replace if in doubt.

Check ambient temperature at the PSU intake with an IR thermometer or a stuck-on thermistor. Target ≤ 30 °C. Every 10 °C above the cap datasheet temperature roughly halves electrolytic life (Arrhenius / `10 °C rule`). A S9 running in a 45 °C garage is aging its PSU about 4× faster than one in a 25 °C basement. Open a window, add an intake fan, duct cooler air, or relocate the miner before continuing.

Meter the `12 V` rail at the PSU output screws under load. Multimeter on DC volts, black probe on a `0 V` screw, red probe on a `12 V` screw. Power up, let the miner reach full hashrate, hold the probes steady for 60 s. Log the lowest reading. Healthy APW3++ on 240 V mains: `12.8-13.4 V` sustained. Marginal: `12.3-12.7 V`. Failing: below `12.0 V`. This single measurement is the foundation of every subsequent diagnostic step — do not skip it.

Meter the rail at the hashboard input connector under the same load. If the delta between PSU output and hashboard input exceeds `0.3 V`, the jumper cable or the voltage-regulating sense cord is dropping voltage under current — cable loss, not PSU loss. Replace the C13/C19 jumper with a fresh heavy-gauge unit and replace the voltage-regulating sense cord with a new part from D-Central's parts inventory. Re-measure to confirm the delta closes.

Swap in a known-good APW3++ / APW7 from another S9, a salvage-grade unit, or a new unit from D-Central's power supply inventory. Repeat Step 6. If the rail recovers on the substitute, the original PSU is degraded — budget for Tier 3 cap refresh or replacement. If the rail still sags on the substitute on a confirmed 240 V circuit, the fault is downstream (hashboards, control board, voltage-regulating path) and DIY is approaching its limit.

Isolate the hashboard load. With a confirmed good PSU, disconnect two of the three hashboards and power up with only one board loaded. Repeat for each board individually. Any board that pulls the rail below `12.6 V` on its own is drawing out of spec — shorted voltage domain, failed PMIC, or dead BM1387 chip creating a short. Move that board to the Tier 3 queue and document which domain (if you have per-domain visibility via DCENT_OS).

Log mains voltage for 24-48 hours with a Kill-A-Watt EZ, a Shelly EM, or an Emporia Vue. Anything sagging below `230 V` on split-phase or below `115 V` on a tolerated-`120 V` install points at the circuit, not the miner. Neighbourhood-transformer peak load (6-10 PM residential) is the most common pattern; utility-level overload requires a service call, not a PSU swap.

Flash DCENT_OS (D-Central's own open-source Antminer firmware) for per-hashboard, per-voltage-domain visibility. DCENT_OS is the Mining Hacker default on S9 — open-source, maintained in public on GitHub, no licensing overhead. Alternatives: Braiins OS+, LuxOS, Vnish. After flashing, let the miner stabilize 20 minutes, then record per-domain voltages live while hashing. Any domain sagging `0.2 V+` below its siblings under the same load has a failing regulator or PMIC on that specific board.

Open the APW3++ / APW7 for a bulk-cap audit (Tier 3 only — advanced). Unplug, wait 5 minutes, then bleed the primary bulk caps through a `10 kΩ / 10 W` resistor to `0 V` reference; these retain `400 V+` for minutes after unplug. Visual pass: bulging tops, crusted rubber plug, brown electrolyte residue = gone. ESR-meter every primary electrolytic with a Peak Atlas ESR70 or Blue ESR; anything above 3× datasheet ESR is drifted and needs replacement.

Replace failed caps with 105 °C / 10,000 h-rated parts. Match voltage rating (typically `450 V` on primary bulk), capacitance, and case size. Nichicon, Rubycon, or Panasonic low-ESR electrolytics only — avoid unknown Asian rebranders, which is how a refreshed APW3++ ends up back on the bench in 18 months. Maintain creepage and clearance for `450 V` primary-side voltages; no shortcuts. Flux-clean all solder joints and inspect for cold joints before closing up.

Inspect and (if needed) replace the PFC MOSFET and boost diode. `RDS(on)` drift is hard to measure without bench equipment, but a visibly discolored TO-247, a cracked solder joint on the MOSFET source lead, or measurably high gate-source leakage is grounds for replacement. Use genuine IRFP460 or equivalent `N-channel 500 V` MOSFETs from Digi-Key or Mouser — no AliExpress counterfeits. Re-paste the heatsink with Arctic MX-6 and a fresh mica insulator.

Post-repair load test the APW3++ before putting it back on a production miner. Dummy-load at `~1200 W` (S9 nameplate) on a resistor bank or a retired hashboard wired as a test load. Hold full load for 30 minutes. Rail stable within `12.8-13.4 V`, PSU case under `55 °C` at the vent, fan RPM appropriate = safe to reinstall. Never skip the load test — a PSU that looks fine at idle can still fail under sustained load if one cap is still marginal.

Stop-and-ship triggers for D-Central Tier 4 repair. Ship to D-Central when: visible cap leak, burn marks, or burnt-PCB smell on the APW3++; a substitute PSU of the correct model on a confirmed 240 V circuit also fails (damage has moved past the PSU); per-domain voltage isolates a regulator you can't replace; you've refreshed caps once and rail sag returned within 90 days; you aren't set up to safely discharge a `400 V+` primary bulk cap. Book at d-central.tech/services/asic-repair/.

D-Central bench process. APW test fixture with programmable AC source — we can replay the exact mains condition that killed your rail to confirm failure mode. Full component-level audit of primary bulk electrolytics, PFC MOSFET, boost diode, secondary rectifiers, control IC, and the voltage-regulating sense circuit. Parts sourced in Canada from authorized distributors (Digi-Key, Mouser, Nichicon) — no counterfeits. 24-hour nameplate-load burn-in at room temperature before return. If hashboards also took damage, per-domain PMIC replacement on the same ticket.

Ship safely to D-Central. Anti-static bag each hashboard individually; bag the PSU separately. Double-box with ≥ 5 cm of foam on every side. Tape the APW3++ AC inlet shut to protect the C20 pins in transit. Include a written note: observed rail voltage idle and under load, firmware version, mains voltage at the outlet, approximate hours of operation since purchase, and your contact details. Better notes = less bench diagnostic time = lower repair cost.