PSU Crowbar OVP Latched — How to Reset

Hard AC-cycle for 60 seconds at the breaker, not the PSU switch. A full breaker cycle drains the bulk capacitor and lets the SCR's holding current fall below threshold so the latch releases cleanly. Rapid power-button toggling does not. Wait the full minute on a watch — premature re-energise re-arms the same trip with a still-charged latch and re-fires the crowbar against the same fault.

Disconnect ALL output cables before re-energising the second time. The crowbar fired because something drove the rail OV. AC-cycling with the same load attached just repeats the fault. Strip the load, AC-cycle, watch the rail. No-load clean = downstream fault on the load side. No-load still dead = PSU internal fault, route to Tier 3+.

Verify ambient environment. If the rig is in a sub-zero garage, warm the PSU to room temperature before retry. APW-class cold-start OV transients on first apply are real — the soft-start ramp overshoots when chemistry is cold. Move the unit indoors for the diagnostic, not just on principle. Re-test at `15-20 °C` ambient.

Inspect every output cable visually before reconnecting. Split insulation at strain reliefs, blackened PCIe bullets, oxidised contact surfaces — any can backfeed or short-trigger the crowbar. Replace, never repair-tape, any cable showing damage. PCIe pigtails running 200 A peak are not a tape-over situation.

If this is the third AC cycle and the rail still isn't recovering, STOP. Either route the unit to bench repair or ohm-test the load before any further AC application. Each additional cycle damages `F1` margin, accumulates SCR gate-junction stress, and increases the chance of permanent damage. Cycle counts above three convert a $95 repair into a $345 rebuild.

Ohm-meter every load output to ground with PSU unplugged. Probe `+12 V` to chassis ground at every PSU output bullet. `>50 Ω` cold = healthy load capacitance. `<2 Ω` = downstream short. `<0.5 Ω` = dead short, almost certainly a FET drain-source weld on the hashboard. Identify which output is the dead one — that cable or hashboard is the suspect.

Disconnect the suspect hashboard and ohm-test it directly. PCIe entry pad `+12 V` to ground; voltage-domain DC-DC stages individually. `Q1`/`Q2` of any domain measuring short drain-to-source means the FET is welded shorted — that domain needs Tier 3+ component-level repair on the hashboard, not the PSU.

Reconnect surviving hashboards one at a time, AC-cycle between each, confirm only one fault. Two faults on one rig is rare but happens after a lightning event — finding the second is the difference between a one-time repair and a return ticket. Document each connection result.

Replace damaged PCIe pigtail cables outright. Never repair-tape a 12 V rail running 200 A peak. Order replacement from the same OEM channel as the PSU. If the original cable is OEM-locked (Whatsminer P21 wiring is a common offender), source the matching part rather than substituting.

Verify line voltage at the outlet under load. If the rig is on a circuit that brownouts under load, the brownout-recovery inrush is what drove the OV transient that fired the crowbar. Plug a high-load resistive heater into the same outlet — if voltage sags below 215 V at the outlet under heavy load, move to a dedicated `240 V / 30 A` minimum circuit before re-deploying.

Tier 3 — open the PSU. Cut AC, wait 5 minutes, discharge the bulk capacitor with a `10 kΩ 5 W` resistor between its terminals. Continuity test across `F1`. Open = blown. Replacement requires the matching slow-blow ceramic with correct I²t rating; typical APW-class is `T20A 250V` or `T25A 250V`. Wrong fuse means either nuisance-tripping or failure to protect on the next event.

Diode-test the crowbar SCR. With AC off and bulk cap discharged, diode-test across anode-cathode. A healthy SCR reads open in both directions; a damaged SCR reads short or low resistance. Common APW-family parts are `BT151`-class or `S6025`-class SCRs. Replace with the same part number — substitutions change the firing characteristics.

ESR-meter the secondary output capacitor bank. Bulging tops, leaked electrolyte at the base, or `ESR > 50 mΩ` on any cap = replace the entire bank. A single shorted cap is what fired the crowbar; the surviving caps are aged similarly and will fail in months. Match cap specs (capacitance, ESR, ripple current rating) and replace as a matched bank.

Inspect the OVP feedback divider resistors and comparator opto-coupler. Drift `>5 %` on either resistor in the divider = replace in matched pairs (1 % precision is the spec). Cracked opto package or burnt board around the opto = replace. These are sub-dollar parts that, if drifted, fire the crowbar at normal rail voltage with no load fault present.

Inspect the secondary rectifier diodes for heat damage. The crowbar event dumps a lot of current through the secondary rectifier in the microseconds before the primary trip catches. Discoloured solder joints, lifted pads, or visibly cooked rectifier packages = replace. Common APW family part: `MBR60100PT`-class Schottky.

Tier 4 — when to stop DIY: `F1` blown twice on the same unit, SCR confirmed shorted, secondary cap bank visibly bulged, opto-coupler cracked, or you don't have line-voltage isolation and an ESR meter. Any of those = ship to D-Central. Component cost is low; the line-voltage risk and AC-side trace-rework difficulty earn this a Tier 4 ticket.

What D-Central does at the bench: full primary-side teardown, `F1` + SCR + secondary cap bank replacement to spec, OVP feedback divider verification with calibrated bench reference, opto-coupler replacement, full re-test from idle through full rated load with thermal imaging on every stage, and a deliberate OV injection to confirm the crowbar fires cleanly and recovers cleanly. 24-hour soak under full miner load before return.

Ship safely. APW-class PSUs are heavy (~5 kg). Original Bitmain box if you have it, otherwise double-cardboard with at least 5 cm of foam on every side. Pull all PCIe cables, coil and zip-tie them separately. Include a note documenting AC cycle count attempted, loads attached, symptoms observed, and any diagnostic readings taken.

Do not ship the load and the PSU together unless both are known-broken. A working hashboard in the same box as a damaged PSU under shipping vibration is a common way to convert a one-component repair into a two-component repair. Keep them separate, label everything, and only ship the parts that need bench attention.

Document the attached load on the repair ticket. `S19 hashboard with bulged domain cap on rail 2` tells our bench tech the SCR fired against a real downstream short and the PSU is otherwise healthy. `Unknown miner, just stopped working` forces full diagnostic re-run. Specific tickets are faster, cheaper, and the unit comes back sooner.