Whatsminer Error 268 – PSU Current Imbalance

Pull per-board telemetry via btminer-api to find the outlier rail. From a laptop on the same subnet: curl http://<miner-ip>:4028 -d '{"cmd":"summary"}', {"cmd":"devs"}', and {"cmd":"status"}'. The devs array shows per-chain MHS av, temperature, and Status. The outlier board will be >5% low on hash (dead chip), >5% high paired with >5C warmer (partial short), or showing one chain in Alive=N. The status response confirms 268 in error_codes. If you cannot reach the API, fall back to LED pattern from the Whatsminer LED reference and proceed as umbrella 268.

Confirm via WhatsMinerTool per-rail PSU telemetry. WhatsMinerTool exposes Power % per output rail in the PSU diagnostic page. Run the miner at tune, log per-rail Power %, V_out, I_out_total, and per-rail temperature for 15-30 minutes. Healthy: each rail within +/-3% of others. Imbalance: one rail at +/-10-25% steady-state. This corroborates the API outlier and rules out a transient. Save the log; if you ship to D-Central, attach to the ticket.

Power-cycle at AC mains, not at the chassis switch. Kill AC at the PDU or wall breaker, wait a full 60 seconds for bulk caps on both PSU and hashboards to discharge. Some BTMiner firmware revisions latch the Code 268 flag through a soft reboot; a full AC cycle clears the latch cleanly. If 268 returns within 5 minutes of resumed hashing, the underlying cause is real and persistent. If it stays clear for >4 hours, you may have caught a transient (rare); log the next event.

Compare board-level OC / UV profiles in WhatsMinerTool. Pull the per-board tune profile and confirm all three boards run the same target frequency, voltage, and chip-bin assumption. If one board was restored from a backup tune that mismatches the others - common after partial-board RMA or firmware downgrade - re-apply a unified tune and observe. Genuine 268 triggered by tune asymmetry clears immediately on re-balance. If tune is uniform and fault persists, asymmetry is hardware-side.

Check service log for recent events on the offending board. Was the suspect board recapped, repaired, chip-swapped, thermal-paste-refreshed, or returned from RMA in the last 30 days? Workmanship-pattern faults - cold solder joint, missed cap, misoriented part - show up as Code 268 24-168 hours after the work. Service log saves the next operator hours of diagnosis. If yes, the board is the suspect and the fix is a re-do plus inspection.

Re-torque the offending board's 12V ring terminal at the busbar. Kill AC, wait 60s, open the chassis. With a calibrated torque screwdriver at 3.0 N-m, loosen 1/8 turn and re-torque each ring-terminal bolt and any M6 busbar bolt on that rail. Look for blackening / blueing of the copper, missing lock washers, or bolts never seated flat. Inspect the cable jacket from busbar to board for heat marks. Re-assemble, run for 1 hour, re-pull devs and status. This single step fixes ~20% of Code 268 returns.

Inspect the 12V cable / harness from PSU to control box to suspect board. Walk the full length under bright light. Look for jacket discolouration or heat marks, visible crimp damage at ring terminals, kinks, pinch points, rodent chew marks, greenish corrosion. Squeeze progressively - soft spots or crunching = internal conductor damage. Pay extra attention to the section feeding the offending board. Genuine MicroBT replacement harness is CAD $30-60. Bad harness presents identically to a bad joint at the comparator level.

Re-flow / re-seat the data ribbon and any SBD adapter on the offending board. A cold solder joint on the chip-temp sensor, the chip-data ribbon, or the SBD (Sub-Board) adapter can cause the PSU's per-rail current sensor to read inconsistent data on that board even when the load itself is symmetric. Pull the data ribbon, inspect connector pins for bent pins or corrosion, re-seat firmly, listen for the click. On M50S+ / M60S+ / M66 class with separate SBD adapter, pull and re-seat that as well. Power up, re-pull devs, observe.

Measure each board's 12V input voltage at the ring terminal under load. With the miner hashing at tune, DMM in DC mode probes at the 12V ring terminal of each of three boards at the chassis side. Healthy: each board within 0.05V of others (typically 12.0-12.2V under load). Imbalance: offending board reads 0.1-0.3V lower or higher than siblings. Consistently lower = high-resistance joint or bad harness; consistently higher = downstream dead chip drawing less current and rail un-loading. Isolates joint problem from load problem without a scope.

Log V/I telemetry for 60 minutes at full load. With WhatsMinerTool running, capture per-rail V, I, Power %, and per-rail temperature at 1 Hz over a full hour. Look for steady-state imbalance (continuous, points at static fault) vs transient imbalance (jitter, points at intermittent contact or chip-switching artifacts). Overlay against ambient temperature and miner workload - a fault that worsens with ambient is thermal (caps drying, heat-soaked PMIC); a fault that worsens with workload is electrical (chip stress, voltage-domain failure). The CSV-export from WhatsMinerTool is a 1-page diagnosis when read correctly.

Back off the tune by one step on the offending board only. If WhatsMinerTool allows per-board tune and your firmware exposes it, drop the offending board's frequency by ~50 MHz and observe. If imbalance clears, the board is genuinely marginal at original tune - bad chip bin, aging silicon, or thermal headroom collapse. Run at reduced tune meanwhile and plan the bench fix. If firmware does not expose per-board tune, drop the whole-miner tune by one step instead - lose a few percent hashrate but tells you whether the fault is tune-related or hardware-related.

Pull the suspect board to the bench and probe each ASIC for low-side shorts. Kill AC, open the chassis, disconnect the data ribbon and 12V ring terminal on the offending board, lift the board to the bench. With DMM in resistance mode (200 ohm range), probe each ASIC chip's 12V input pin to ground. Healthy chip reads >1 kohm. Partial short reads 100-500 ohm. Hard short reads <10 ohm. The bad chip will be obvious - single chip out of ~150 reading <500 ohm. Mark its position. Confirms category 1 (dead/shorted chip).

ESR-test the 12V bulk capacitors on the offending board. With the board on the bench and an ESR meter (Atlas ESR70, MK-328, or equivalent), probe each 12V bulk electrolytic. Healthy reads <50 mohm ESR. Aged-but-acceptable reads 50-100 mohm. Failed reads >200 mohm and is the failure-category-2 culprit. Visual sanity check: any cap with top-bulge, electrolyte weep at the leg, or discolouration around the body should be replaced regardless of ESR reading. Recap kit specs: same capacitance, equal-or-higher voltage rating, low-ESR series (Panasonic FR / Nichicon HE / Rubycon ZLH), 105C-rated, same physical footprint.

Reflow or replace the bad ASIC chip. If Step 12 isolated a partially-shorted or dead chip, remove the heatsink, clean with 99% IPA, flux the BGA underside, preheat the bottom side to ~150C, apply hot air at 310-330C from the top for ~30 seconds while observing under shop light. For reflow attempt: cool slowly, re-paste, re-assemble, re-test. For replace: remove the chip with hot air, clean pads, ball or paste a new chip from a graded-salvage source, re-flow at same temps. Whatsminer ASIC chips are a pull-and-replace job that reflowable solder paste handles cleanly.

Re-cap the offending board if Step 13 failed any caps. Mark cap orientation on silkscreen with a marker before removal (electrolytics are polarity-critical). De-solder failed caps with hot air or dual-iron technique, clean pads with desolder braid + IPA, install new caps with correct polarity, flux + reflow, verify with multimeter for proper polarity post-install. Re-assemble the board into chassis, power up, run WhatsMinerTool PSU telemetry for 30 minutes to confirm Power % symmetry. Recap fixes category-2 cleanly; the board typically runs 2-4 more years before next cap event.

Replace the per-rail sense resistor on the PSU board (skill gate, sensor-circuit failure case). If Step 7 of diagnostics confirmed the PSU's per-rail sensor is lying (Code 268 follows the PSU to a known-good chassis), open the PSU chassis (kill AC, wait 5 minutes for primary-side caps to discharge), locate the per-rail shunt resistors (1 mohm, 0.5 mohm, or 2 mohm depending on PSU revision - silkscreen has the value), inspect under magnification for hairline cracks. Replace with same-value precision shunt (Vishay / Bourns / Susumu). May need to replace the differential amplifier IC adjacent to the shunt - same hot-air SMD rework. Skill gate: if you cannot do 0805 / 0603 SMD rework with hot air, ship to D-Central.

Check grounding / bonding between PSU chassis and control-box chassis. On rare deployments - typically container farms or multi-miner power racks - a poor bond or ground-loop between the PSU chassis and control-box chassis creates a measurable >100 mV potential difference under load, couples into the per-rail current sense circuit, and presents as Code 268. Measure with DMM between PSU chassis and control-box chassis under full load: <50 mV is fine, >100 mV is a real ground problem. Investigate the earth path from PDU through to building ground.

Bench-test the suspect PSU against a programmable DC load. If you have access to a DC electronic load rated >400A @ 12V (TDI / Chroma / Itech-class) capable of three independent channels (or single channel cycled across rails), pull the PSU and run a per-rail balanced load profile: 100A on rail 1, 100A on rail 2, 100A on rail 3, all simultaneous. Log per-rail V and I reported by the PSU vs the load's measurement. Discrepancy >5% between PSU-reported and load-measured = sensor-circuit fault inside the PSU. Documents the fault for warranty / RMA or confirms replace-vs-repair decision.

Stop DIY when Tiers 1-3 are clean but Code 268 still latches within hours. The fault is likely combined - a hashboard with early-stage chip silicon failure that only reveals itself at full duty cycle, or a PSU sensor circuit drifting on a healthy-but-marginal load curve, or both. Bench isolation with a programmable DC load on the PSU AND a hashboard fixture checking each board at rated current is the only reliable path. Home gear cannot reproduce; chasing with reboots wastes operator hours and accelerates degraded silicon. D-Central's bench process pins the category in one afternoon.

Stop DIY when you see scorched silicon, blued copper, or cracked packages on the offending board. Scorched buck-regulator MOSFETs, blued copper at any joint (joint hit >250C), cracked package on a PMIC or ASIC, or burnt silkscreen are all signals of cascade damage. Replacing only the visibly-broken part leaves the miner vulnerable - adjacent damaged-but-not-yet-failed parts are next. D-Central's bench repair recaps the full 12V chain, replaces damaged silicon, and inspects adjacent components under magnification. Cascade damage often costs the same to repair as a graded-salvage replacement; diagnostic call is photo-documented up front.

Ship with full diagnostic context. Pack the chassis with the PSU (we need your exact stack to reproduce the per-rail imbalance - cross-stack substitution misses subtle interactions), a copy of your last btminer-api summary + devs + status JSON, the WhatsMinerTool PSU telemetry log, photos of any visual anomaly, service history (recent moves, board swaps, recaps, tune changes), and intake-ambient log. Every minute of context saved is a minute of turnaround shortened. Match chassis serial to PSU serial in your ship note. Canada-wide standard shipping; US / international welcomed.

Discuss repair-vs-replace for the offending board. A bench-level chip-replace + recap on a Whatsminer hashboard runs CAD $120-280 depending on what failed. Graded-salvage replacement runs CAD $300-700 depending on model. If the offending board is 5+ years old AND has multiple dried caps AND a bad chip, math often points at replacement. If <3 years old and fault is one chip + clean caps, repair almost always wins. D-Central quotes both options up front with photo-documented diagnostic. For PSU sensor-circuit failures, replace usually wins - PSU silicon ages on a coherent curve.