Antminer S21 – PSU Overcurrent Protection

Full mains cold-cycle at the breaker for `30 s`. Not a UI reboot — actually kill the circuit for 30 seconds so the 400 V bulk caps discharge enough for the supervisor IC to fully reset the OCP latch. Some APW21 firmware revisions keep OCP latched across a short power blip; 30 s at the breaker clears it reliably. Re-power. If the miner refuses to boot, jump to Step 2. If it boots but trips again within 30 minutes, continue through Tier 1.

Revert to stock Bitmain profile with no OC. Disable any tuning, autotune, voltage-scaling preset, or third-party firmware overclock config. Target nameplate `3531 W` at the wall for standard S21 (200 TH/s), `3538 W` for S21 Pro 234 TH/s bins. Run 30 minutes at full hash, confirm no ERR_OCP recurrence. If stable, tune back up +50 MHz at a time with 10 min stability between steps, stop before OCP returns. That last stable step is your safe ceiling on this PSU + chip + ambient combination.

Vacuum the APW21 intake, miner chassis intakes, and the gap between them. Dust loads PSU internal temperature and drifts the OCP current-sense reference closer to your operating current. On the S21 layout the PSU sits closer to hashboard 2 exhaust than on the S19, so intake dust matters more. Shop-vac the PSU grille, canned duster through the fin stack with the PSU unplugged, same on chassis intakes. Monthly on home install, weekly in a dusty workshop.

Confirm intake ambient temperature is `<= 30 C` measured `5 cm` from the PSU intake grille with an IR thermometer — intake-specific, not room-middle or HVAC register. The S21's thermal design margin is tighter than the S19's; above `35 C` intake, the PSU's supervisor sees enough temperature drift that OCP trips cluster on afternoon peaks. If ambient is too high, add a box fan across the intake, reorient, or relocate the miner before escalating.

Verify firmware version against the Bitmain S21 downloads page at `support.bitmain.com/downloads`. The earliest S21 shipping Bitmain firmware has a documented PMBus handshake timeout that presents as spurious ERR_OCP on cold boot. Update to the latest Bitmain build, or cross-flash DCENT_OS (Tier 3 Step 11) which patches the handshake path entirely and logs handshake state so you can diagnose failures. Keep a known-good firmware image on a prepared SD card before you flash.

Measure mains input voltage under full load. Multimeter on AC at the PSU `C20` inlet, or a Kill-A-Watt / Sonoff POW R2 inline. Band targets: `235-245 V` on 240 V split-phase residential, `225-235 V` on international 230 V, `202-212 V` on 208 V commercial. Log `24 hours` to catch evening sag. Below band is the root cause — sag forces the PSU to pull more input current to make the same output wattage, OCP fires on the excursion. Document the worst sag and timestamp.

Measure DC output voltage under load. DMM on DC at the APW21 output lugs while the miner is at full hash. Healthy APW21 reads `13.8-15.1 V` under load (varies because the rail is dynamically negotiated via PMBus), `14.8-15.0 V` at idle. Sustained `< 13.5 V` under load while the control board commands higher = PSU is sagging internally, bulk-cap aging confirmed, Tier 3 territory. Record the worst-case reading and the commanded frequency at that moment.

Inspect and reseat the PMBus signal cable. Power off at breaker. Disconnect the PSU signal cable at both ends — APW21 side and control-board side. Inspect with a bright light for oxidized pins, bent contacts, cracked housing, or loose strain relief. Clean with DeoxIT D5, work the connectors 5-10 insertion cycles to wipe mating surfaces, reseat firmly. This is the most under-diagnosed S21 OCP cause — a flaky handshake makes the APW21 default to a fallback voltage the hashboards can't chain-boot against.

Swap hashboards between slots. Label slots `0/1/2` with tape. Move the suspect board to a known-good slot, boot, measure per-chain current on a clamp meter at the DC output lug. Per-board expected draw: `95-110 A` at nameplate standard S21, `110-120 A` for S21 Pro 234 TH/s. If high current follows the board = bad board, tag for Tier 3 diagnostics. If it stays with the slot = bad control-board feed, PDB trace, or inter-board loom. Either way, cause isolated in under `15 min`.

Move the miner to a dedicated `240 V / 20-30 A` circuit. If your S21 is on anything shared — a dryer circuit, EV charger, baseboard heater, or a PDU with other miners — move it. The APW21 is `200-240 V` only and pulls `15-18 A` sustained. Shared circuits sag during load transients and drive OCP trips. Plan `$400-800` CAD for a Canadian electrician. On a single-S21 home install this pays back in `12-18 months` via avoided PSU damage alone.

Flash DCENT_OS (D-Central's own open-source Antminer firmware, the Mining Hackers' option, preferred) for per-chip, per-chain, per-rail, and PMBus input telemetry that stock Bitmain hides. Source on GitHub. Alternatives: Braiins OS+, LuxOS, Vnish. Graph PSU input current and per-chain current for `48 hours`. If excursion patterns match the OCP trips 1:1, the miner is the cause. If they don't, the PSU's OCP reference has drifted from age or heat — Tier 4 bench territory.

DANGER — discharge the bulk capacitors before opening the APW21. `400+ V DC` lives on the primary bulk caps for `30+ s` after unplug, longer on S21-era units with larger reservoir caps. Short each cap through a `1 kOhm / 5 W` resistor for a full `10 s`, verify `< 5 V DC` with a DMM before any tool touches metal. This is not optional theatre — 400 V across your chest kills you. If not 100% confident in this procedure, stop here and ship the PSU to D-Central (Step 16).

ESR-audit every electrolytic on the APW21 with a Peak Atlas ESR70 or DE-5000 in-circuit. Any cap `>30%` above stock ESR is replacement territory. Priority order: primary-side bulk caps (verify footprint against your revision — typical `330 uF / 450 V` pair), LLC output filter caps on the `12-15 V` rail, auxiliary rails feeding the supervisor IC. Replace with `105 C` minimum, `135 C` if footprint matches. Full cap + paste + fan refurbish buys `3-5 years` additional life for `$25-55` in parts.

Re-paste APW21 primary-side heatsinks after discharge. Remove top shell (fastener count varies by revision — count yours before disassembly). Clean old paste with `99%` IPA off the PFC diodes, primary MOSFETs, and synchronous-rectification MOSFETs on the secondary side. Apply Arctic MX-6 or Kryonaut in a thin uniform layer, reassemble with even torque. Dry paste on a two-year APW21 can raise primary junction temperature `10-15 C` and drift the OCP reference — explains the month-20 failure cluster we see at the bench.

Replace the APW21 internal fan if it whines, ramps late, or doesn't ramp at all. Stock on most APW21 revisions is a `120x120x25 mm` or `120x120x38 mm` axial, `12 V`, dual-ball-bearing, `1.5-2.0 A` rated. Note the 3-pin or 4-pin connector before pulling — the RPM-tach wire is read by the supervisor IC and a 2-pin replacement will throw a fan fault that can trigger a protective pre-OCP shutdown on some firmware revisions that cross-wire the two protections.

Stop DIY and book a D-Central slot when: (a) OCP reproduces with a known-good substitute APW21 (miner-side fault, test-fixture territory); (b) visible cap bulging, vented rubber plug, discolored PCB, brown residue, or burnt/ozone smell; (c) ESR audit finds `>3` electrolytics above spec or any primary MOSFET reads short on diode check; (d) per-chip current isolates a failing chip position on two different boards (PMIC / domain-IC work); (e) you are not 100% confident in the Step 12 discharge procedure. Book at `https://d-central.tech/services/asic-repair/`.

What D-Central does at the bench: APW21-specific test fixture with programmable electronic load to `300 A` at `12-15 V`, full PMBus handshake emulation to verify voltage-setpoint negotiation without a live miner, thermal chamber (`25 / 40 / 55 C` ambient) for OCP threshold drift characterization, full ESR + capacitance + leakage audit on every electrolytic, primary-side semiconductor check, selective re-cap, full re-paste, fan replacement if needed, and `24-hour` full-load burn-in at nameplate before ship-back. S21 hashboards get per-chain current profiling on the same bench.

Ship safely. APW21 in its own anti-static bag, wrapped separately from the miner. Double-box with `>=5 cm` of foam on every side — bulk caps are heavy, they punch holes in single-box packs. Include a written note with observed symptoms, firmware version and build date, exact `ERR_OCP` / `ERROR_POWER_LOST` timestamps from `kern.log`, measured rail + mains + ambient values, whether a substitute PSU reproduced the trip, and your contact. Five minutes of notes saves `30-45 min` of bench re-diagnostic — real dollars off your ticket.