Whatsminer Hydro – Water Leak Alarm Trip

Emergency kill at the PDU. Trip the breaker or pull the C19 at the PDU — do not use the web-UI shutdown or the chassis power switch. A leaking hydro server with the PSU still energized is a chip-graveyard in slow motion; every second of live silicon against coolant kills another chip. Wait 5 minutes after power-off to let coolant pressure bleed off and PSU bulk capacitors discharge before any physical inspection.

Visual-only leak triage with power off. Shine a flashlight at every external surface: chassis floor, drip tray, quick-disconnect couplers, CDU return line, reservoir area. Photograph anything wet, discoloured, or crusty. Note residue colour: clear = fresh weep, white crust = evaporated weep with mineral deposits, green tint = glycol additive. Do not attempt any fix at this tier — document for the bench phase.

Reservoir level check against operator logs. Read the CDU / bulk reservoir level mark and compare to last week's reading. A drop of ≥ 100 mL per week on a typical 4-10 L Whatsminer hydro loop with no scheduled top-ups confirms a real leak. No measurable drop points toward a false-positive sensor trip — proceed to sensor-clean in Step 10.

Environmental / dew-point check. Measure room temperature and relative humidity at the server intake with a Fluke 971 or equivalent. Compute dew point: Td ≈ T − ((100 − RH) / 5), in °C. If coolant supply temperature is below Td, the system is condensing atmospheric moisture onto hose runs and QD bodies — this reads as a leak to the sensor but is an environmental problem, not a hardware leak.

Do NOT power-cycle repeatedly to clear the alarm. MicroBT firmware latches `WATER_LEAK` until the sensor reads dry. Each power cycle re-pressurizes the loop against a compromised seal, accelerating the failure. Leave power off until you have isolated and fixed the root cause.

Drain and remove the server from the coolant loop. Close isolation valves if your farm has them, uncouple QDs over a drip tray or 10 L bucket, expect 0.5-1.5 L of back-drained coolant per server. Cap both QD ports with supplied dust caps. Move the server to a dry bench with drip-tray coverage. Dry the exterior with lint-free cloth.

Low-pressure compressed-air test. With QDs capped on one side, connect low-pressure air (0.10 MPa / ~15 PSI maximum — do NOT exceed nameplate) to the coolant inlet. Submerge each QD coupler in a shallow water tray and observe for bubble streams. This is the field version of the fluorescein test and catches approximately 70% of leaks without dye. Mark any bubbling fittings for replacement.

Coolant chemistry sample analysis. Draw 10 mL of coolant from the reservoir. Check: clarity (cloudy = biological growth), colour (rust-tinged = internal corrosion, inhibitor depletion), pH (should read 7-9 on most hydro coolants — acidic = chemistry failure). Degraded coolant chemically attacks O-rings from the inside; a server that was leak-free for 18 months then suddenly starts leaking at every QD typically indicates coolant chemistry failure, not mechanical damage.

Walk the operator-side loop. Inspect every hose clamp, fitting, elbow, and hose run from the CDU / dry cooler back to the server. A leak outside the server can pool at a low point, wick back through QDs, and trip a server-side sensor that is working correctly. Fix upstream leaks and the downstream alarm self-clears. Zeus BTC's hydro repair guide has diagrams of typical CDU failure points.

Sensor-pad clean and fault-register reset. Open the bottom cover, locate the two-pad moisture probe wired to the control board. Clean pads with 99% isopropyl alcohol and a small nylon brush — never use cotton swabs, as fibres bridge the pads and create a new false positive. Air-dry for 15 minutes. Reassemble. Reset the fault via `btminer-cli fault-clear` over SSH or WhatsMinerTool's 'Reset Alarms' button. Watch the log for 5 minutes to confirm the alarm does not re-arm.

Fluorescein UV pressure test (the professional bench technique). Drain, add 5-10 ppm of food-grade fluorescein dye to the loop, re-pressurize the server-only section to nameplate (typically 0.15-0.25 MPa — verify against your model's spec sheet), hold for 15 minutes. Inspect the entire server under a 365 nm UV lamp. Fluorescein localizes the leak to a single fitting, gasket edge, or weld line. This technique is standard in automotive A/C and industrial hydraulic repair; a complete kit is approximately $40 CAD.

Quick-disconnect O-ring replacement. Remove the failing QD. Extract the primary face seal O-ring and secondary radial O-ring. Replace with FKM / Viton 75A O-rings rated for glycol exposure and 0-150 °C service. Lubricate with food-grade silicone O-ring grease (Super Lube 92003 or equivalent — never petroleum-based grease, which swells FKM and causes secondary failures). Re-torque the QD body per MicroBT's hydro service manual — typically 8-12 N·m, but verify per model with a calibrated torque wrench.

Coldplate gasket replacement (advanced, chassis-teardown). Only attempt if the fluorescein test localized a weep to the hashboard-coldplate joint. Sequence: disconnect all harnesses, remove the 12-bolt coldplate pattern in star-torque reverse order, lift coldplate clear of the hashboard, inspect gasket for extrusion / tears / pinch points, replace with MicroBT OEM silicone gasket (do not substitute — thickness and durometer are critical), reassemble in star-torque sequence to specification (typically 3-5 N·m per bolt — verify per model). Get any step wrong and the server will leak again within 30 days.

Moisture-sensor swap (if physically damaged). If the sensor shows corroded pads, a broken probe body, or a severed harness, replacement is straightforward: four-wire keyed connector, approximately $15 CAD from MicroBT-authorized resellers. Beware of third-party clone sensors that report wet continuously — stick to OEM or verified known-good pulls. Install, route harness away from coolant path, and re-test in dry state before re-connecting to the loop.

Coolant replacement. If Step 8 showed degraded coolant chemistry: drain the entire loop, flush with distilled water until the discharge runs clear, refill with MicroBT-spec coolant (propylene-glycol / distilled-water blend at the concentration specified for your model, with corrosion inhibitor). Never use tap water (mineral content) or automotive antifreeze (wrong inhibitor package, attacks seals). Document the refill date and set a 12-month replacement interval in your maintenance calendar.

Stop DIY and ship to D-Central when: the fluorescein test isolates the leak to the coldplate gasket and you lack a calibrated torque wrench, coolant residue is visible on the hashboard PCB itself, multiple chips were submerged while the server was energized, or the chassis shows heat damage (burnt smell, discoloured PCB, desoldered components) near the leak path. These conditions require bench-level diagnostics and repair capability.

D-Central bench process for hydro leak repairs: full teardown, board-level fluorescein retest at operating temperature, coldplate re-gasketing to OEM specification with calibrated torque, coolant-exposed chip inspection under microscope, PMIC and voltage-domain verification under programmable load, per-chip SHA-256d validation on the test fixture, re-seal, 24-hour burn-in at nameplate hashrate. Every coldplate bolt torque value is recorded in the service record. The loop is pressure-tested one final time before the server leaves the bench.

Shipping preparation. Drain the server completely. Blow out residual coolant with low-pressure compressed air (0.05 MPa maximum). Cap both QDs with dust caps. Double-box the server with at least 5 cm of foam on every side. NEVER ship a hydro server with coolant in it — transit vibration can damage hoses and turn a repairable leak into a chassis-wide flood. Label the outer box 'HYDRO ASIC — DRAINED — FRAGILE'. Include a diagnostic note listing symptoms observed, log excerpts, firmware version, coolant age, and recent maintenance history.