Whatsminer Error 233-238 – PSU Output Protection Triggered

Confirm the exact sub-code via the BTMiner API, not just the Web UI. Curl port 4028 with {"cmd":"status"} from a laptop on the same subnet and parse the error_codes array. [233] is the umbrella fault; [234] / [237] point at overcurrent; [235] points at overvoltage / input sag; [236] points at a fast short-trip. This distinction drives everything downstream. If you cannot reach the API (network down, control board unresponsive), fall back to the LED pattern from the Whatsminer LED reference and proceed as umbrella 233.

Power-cycle at the AC mains, not the chassis switch. Kill AC at the PDU or wall breaker, wait a full 60 seconds for primary-side bulk caps and secondary-side output caps to discharge. A short chassis-button reboot leaves the PSU MCU's fault latch set on some firmware revisions and the error re-appears instantly. A full AC cycle resets the latch cleanly. If the error clears on AC cycle and returns within minutes under load, you have a real fault, not a stuck flag.

Check the last 24-72h for physical or electrical events. Was the miner moved, bumped, had its PSU swapped, had a lid off? Was there a thunderstorm, a brownout, a neighbourhood grid event, a breaker trip-and-reset? Recent physical events point at loose hardware (busbars, harness, ring terminals). Recent electrical events point at PSU damage (OVP ref drift, MOSFET degradation, sense-resistor crack). Service log saves hours of rediagnosis on the next event.

Inspect intake airflow and ambient. Pop the intake grille, look for dust mats, check fan RPM on the dashboard (>=3,500 RPM on M30S class, >=5,500 RPM on M50S+ class). Intake >35C or restricted airflow raises internal PSU temperature and narrows the OVP / OCP margin. Ambient >40C at the grille almost guarantees sporadic Code 233 / 275 events on an M50S++ or M60S+ at full tune. Canadian winter mining: basically never your problem from November to April - one reason the cold-climate angle matters.

Back off the tune / overclock profile by one step. If Code 233 started appearing within days of pushing the power-mode higher (e.g., 3.4 kW -> 3.8 kW on an M30S++), revert to the previous profile via WhatsMinerTool or the Web UI. Marginal-contact busbars and aged PSUs both lose their safety margin first at the highest power mode. A 10-15% power reduction often clears the fault without any hardware work - confirming the underlying system is near its limit even if not obviously broken yet. Plan the hardware fix for when you have time; run at the reduced tune meanwhile.

Measure AC input voltage under boot load with a clamp meter. Clamp on L-L at the C19 inlet, cold-start the miner, watch the voltage during the first 60 seconds of boot and ramp-up. 240V nominal should stay >225V through the inrush; 208V Wye should stay >200V. If you sag below those, your electrical feeder is undersized for the miner's peak draw and the PSU is seeing conditions it legitimately should trip on. Fix: dedicated 240V 30A circuit, #10 AWG wiring on runs >8m, no shared loads, ideally a PDU that logs voltage so you catch the intermittent events later.

Torque every M6 copper-bar bolt to spec. Kill AC, wait 60s, remove the PSU-side cover and the control-box cover. With a calibrated torque screwdriver set to 3.0 N-m, check every M6 copper-bar bolt on the 12V output (typically 4 at the PSU, 4 at the control-box input). Loosen each 1/8 turn, then re-torque to 3.0 N-m (MicroBT spec is 2.5-3.5 N-m per the self-service guide). Look for blackened / blued copper at the joint (overheated), missing lock washers, or bolts that were never seated flat. Re-assemble, power up, hash for an hour, re-check. This single step fixes ~65% of Code 233 returns per D-Central's repair queue.

Visually inspect the 12V output harness end-to-end. On remote-PSU chassis (most M30S-class), the 12V cable between PSU and control-box is a wear item. Under bright light, walk the full length looking for jacket discolouration or heat marks (prior current overload), visible crimp damage at the ring terminals, kinks, pinch points, rodent chew marks, greenish corrosion. Squeeze the cable progressively - any soft spots or crunching = internal conductor damage. Genuine MicroBT replacement harness is CAD $30-60, cheaper than the hour you'll waste on a bad cable.

Check the control-box side of the busbars. Most DIYers only torque the PSU side. The control-box side - where the output harness terminates and feeds the hashboards - matters just as much and has just as many bolts. Same procedure: 1/8 turn loose, re-torque to 3.0 N-m, look for heat damage. A loose control-box-side joint presents electrically identically to a loose PSU-side joint, and both hide from a one-side-only inspection.

Log V/I telemetry via WhatsMinerTool for 30-60 minutes under load. WhatsMinerTool exposes live PSU telemetry the Web UI hides. With the miner running at tune, log V_out, I_out, P_out, and PSU-internal temp. Healthy baseline: V_out steady at 12.0-12.3V with <0.1V ripple, I_out stable within +/-3% of expected for your tune, P_out matching the dashboard. Precursors: V_out drooping >0.3V under load (loose joint or aged PSU), I_out oscillating +/-10% (flaky contact), I_out brief spike before a Code 233 latch (short-circuit or PMIC failure).

Reduce restricted-airflow environmental factors. If the miner is in a closet, garage corner, or rack with <30 cm intake clearance, fix that. PSU internal temperature rises non-linearly with restricted airflow, and Code 233 / 275 pairs become common above 70C internal. Pull the miner into open air, run for a day, see if fault frequency drops. Permanent fix is proper ducting (D-Central stocks universal ASIC shrouds and duct adapters) or an in-home Bitcoin Space Heater configuration that uses the miner's heat as home heat instead of fighting it with A/C.

Boot into the BTMiner recovery / SD card mode and re-check the PSU firmware. On a Code 233 that persists through torque, harness swap, and input-voltage fix, re-flash the control-board firmware from an SD card via the WhatsMiner bootloader recovery procedure. A corrupted PSU calibration blob in flash occasionally mis-sets the OVP / OCP references and triggers Code 235 on a healthy PSU. Factory-fresh firmware flash clears this. VERIFY the firmware image matches your exact model and sub-revision - cross-flashing an M50S image onto an M50 bricks the control board because MicroBT version-checks on the CB identity EEPROM.

Pull hashboards one at a time to isolate a rail-dragger. Kill AC, open the control-box, disconnect all three hashboards from the 12V busbar - ring terminals and data ribbons both. Re-apply AC. PSU boots cleanly with no fault = PSU is healthy, the trip is downstream. Re-connect one board at a time and power up between each: all three boards clean = no board is dragging; adding board N triggers the fault = board N has a shorted PMIC / failed buck regulator / popped capacitor pulling 200A+ transient. This is the definitive isolation between 'bad PSU' and 'bad hashboard being blamed on PSU'.

Thermal scan under load. With a FLIR One or equivalent IR camera, scan the PSU output busbars, the output harness end-to-end, the ring terminals, and the control-box input while the miner is hashing at full tune. Healthy joints run 35-50C above ambient evenly along the busbar length. Unhealthy busbar = a single hot spot at the loose bolt, often >80C - visually obvious on the thermal image. This visual confirmation before a re-torque saves the trial-and-error of torque-every-bolt-and-hope.

Inspect hashboard PMICs and buck converters under magnification. If Step 13 isolated a specific hashboard as the rail-dragger, pull the board and inspect the 12V-to-core-voltage buck regulators and any PMIC IC under 10x-20x magnification. Blown buck = scorched package, lifted pads, or visible cracked silicon. Shorted output capacitor = swollen top, discoloured solder joints, or a visible burn mark. If you have hot-air + preheater + SMD replacement parts, the board is repairable at the component level. If not, ship to D-Central for bench rework - per-chip repair runs CAD $100-250 depending on what's failed.

Replace a sense-resistor or OVP-reference component on the PSU board (skill gate). A rare but real failure: a 1 mOhm sense resistor on the output-current path cracks internally, drifts high in resistance, and the PSU over-reads the current - firing Code 234 spuriously on a normal load. Visual under magnification: hairline crack in the resistor body or browning around it. Requires hot-air SMD rework and an exact part-number match from silkscreen. OVP reference components (shunt Zener, precision reference IC) fail similarly but rarer. If you have the skill set, fine; if not, skip to Tier 4.

Check grounding / bonding between PSU chassis and control-box chassis. On some deployment configurations (mining container farms especially), a ground-loop or poor bond between PSU chassis and control-box chassis creates a measurable potential difference that interferes with the PSU's output-current sense circuit. Measure with a DMM between PSU chassis and control-box chassis under full load: >100 mV = ground loop, investigate the earth path from PDU through to building ground. Rare at home-miner scale but real in container farms running multiple M50S++ at 4.5 kW each.

Test the PSU against a DC electronic load (bench-only). If you have access to a programmable DC load rated >400A @ 12V (TDI / Chroma / Itech-class), pull the PSU from the chassis and test it on the bench with a load profile that mimics the miner's duty cycle: 50% of rated for 1 minute, 100% for 1 minute, 0% for 10 seconds, repeat. A marginal PSU will trip its protections under the programmable profile even though it appears healthy under static bench load. Documents the fault for warranty / RMA if applicable; confirms replacement vs keep-as-spare decision if out of warranty.

Stop DIY when Tiers 1-3 are clean but Code 233 still latches within hours. The fault is likely combined - a PSU with a drifting sensor reading a healthy-but-marginal load, or a hashboard with an early-stage PMIC failure that only reveals itself at full duty cycle. Bench isolation with a programmable load on the PSU AND a hashboard fixture checking each board at rated current is the only reliable path. Home gear cannot reproduce the conditions. D-Central's bench process pins the fault category in one afternoon instead of a weekend of chasing.

Stop DIY when you see scorched silicon, blued copper, or cracked packages. Scorched output MOSFETs on the PSU board, blued copper at a busbar (joint hit >250C), cracked electrolytic capacitor tops on the hashboards, or visible burn marks are all signals of cascade damage - replacing only the visibly-broken part leaves the miner vulnerable to the next event because the nearby damaged-but-not-yet-failed parts are next. D-Central's bench repair recaps the full input-protection chain, replaces damaged output MOSFETs, and inspects adjacent silicon under magnification - not the visibly-burnt part in isolation.

Ship with full context. Pack the chassis with the PSU (we need your exact stack to reproduce the fault - cross-stack substitution will miss marginal interactions), a copy of your last btminer-api status JSON, the WhatsMinerTool PSU telemetry log if captured, service history (recent moves, PSU swaps, tune changes), and your intake-ambient log if you have one. Every minute of context saved is a minute of turnaround shortened. Match chassis serial to PSU serial in your ship note - both are on metal plates, don't guess. Canada-wide standard shipping; US / international welcomed.

Discuss repair-vs-replace for the PSU specifically. A brand-new MicroBT P21 runs CAD $220-320, a P221 $340-420, a P222 $420-540, and graded-salvage units run ~60% of those numbers at D-Central. A bench-level comm / sense / MOSFET repair runs CAD $80-180 depending on what's failed. If the PSU is 5+ years old and the output-side MOSFETs are beat up, repair might be false economy - the rest of the unit is on the same aging curve. D-Central quotes up front with a photo-documented diagnostic report so you can choose. For hashboard-side faults blamed on the PSU, bench repair almost always wins - hashboards are the expensive part to replace.