A water-cooled hydro Antminer (S19 XP/Pro+ Hydro, S21/S21 XP Hydro) is diagnosed in layers, from the outside in: read the front-panel indicator LEDs, watch the run-state and coolant loop, inspect the web dashboard, then read the kernel log. Each layer narrows the fault from “something is wrong” to a specific board, hydraulic problem or PSU failure.
Why a hydro unit is different
A hydro Antminer is fanless: no fans to check, no filters to clean, no fan-tach faults. Heat is pulled off the hashboards by cold plates bolted to the chips and carried away by coolant. The unit is one node on a closed loop — inlet quick-connect → internal cold plates → outlet → external coolant distribution unit (CDU) or dry cooler. So a large share of hydro faults are hydraulic, not electronic: no flow, a closed valve, coolant over temperature, or a trapped air pocket. Do not apply air-miner steps (fan swaps, filter cleaning) here — check the water side first.
Two safety facts. The low-voltage DC bus-bar feeding the boards (~12–21 V) is not a shock hazard. The danger is the PSU primary side, where mains and PFC capacitors hold ~410–420 V for minutes after power-off. Hydro units run on an APW12-class supply that is 200–240 V only — no 110/120 V mode. Kill mains, wait, and confirm the bus is dead before touching the PSU.
Layer 1 — Read the indicator LEDs
The front panel carries three signals — the Ethernet port LEDs (link + activity), a red Fault LED and a green Normal LED. A fast first read:
- No Ethernet link light → network layer. Re-seat the patch cable both ends, try a known-good cable, confirm the switch port is up. A dead ETH LED on an otherwise-running miner means cable, port or switch — not the control board.
- Red Fault LED on / flashing → firmware tripped a protection. On a hydro unit the usual cause is coolant over-temperature or low flow: confirm the inlet and outlet interface valves are fully open, the CDU pump is running, and coolant temp is within the unit’s inlet-temp spec. Clear the hydraulic fault before suspecting the boards.
- Green Normal LED off, powered but no IP → possible control-board hang. Power-cycle, then hold Reset to force a re-boot / DHCP request. Still no IP after a clean reset → the control board is the prime suspect.
- Link present but unreachable → IP conflict or wrong subnet; see Layer 3.
Layer 2 — Observe run-state and the coolant loop
- Cycles on then drops out / reboots → unstable input. A hydro unit demands full 200–240 V; under-voltage, an undersized circuit, or a loose bus-bar bolt / PSU output lead browns it out under load. Torque-check the DC bus-bar, inspect the AC feed; repeated hard drops also point to a failing PSU.
- Runs but hashrate is well below rated → a board is throttling or partly dead; a single board at zero or half share drags the total down. Confirm coolant is actually reaching that board (flow, not just a running pump) before condemning silicon.
- One board hot, others cool → a starved flow path or trapped air pocket; purge/bleed the loop per the CDU procedure.
Inspect the quick-connects and cold-plate unions for coolant weeping. The low-voltage bus-bar makes a small seep no shock risk, but coolant on a live board is a short and corrosion risk. Any active leak — power down and re-seat or replace the fitting/O-ring before running.
Layer 3 — Recover a lost IP
If the miner is powered and linked but unreachable:
- Use Bitmain’s IP Reporter: run the tool on a PC on the same LAN, then briefly press the miner’s IP-Report button — the miner answers the broadcast with its current IP (the miner replies, it is not a network scan).
- Or check the router / switch ARP table, which resolves each live IP to its MAC; match the Bitmain MAC prefix to find the unit.
- If it never appears, hold Reset to restore DHCP defaults, then re-scan. Still nothing after a clean reset → treat the control board as faulty.
Layer 4 — Inspect the web dashboard
Open the miner’s IP and go to the Miner Status page — where a fuzzy symptom becomes a specific board:
- A hashboard reading 0 GH/s (board 2 flat while 1 and 3 hash) → that board dropped off the chain — a chip-comm break, a power-domain fault, or a coolant-starved board that thermal-tripped.
- Fewer chips found than expected → the ASIC chain is broken. Hydro chips power in voltage domains — a group of chips per domain (e.g. BM1366 on the S19 XP Hydro, BM1368 on the S21 Hydro). Voltage is regulated per domain, not per chip, so one failed domain silences its whole block of chips, not a single chip.
Layer 5 — Read the kernel log
When the dashboard is inconclusive, open the Kernel Log from the miner’s web menu. It records each boot and run stage and usually names the fault outright:
- Network / pool errors → on pool or link loss the firmware idles the miner, it does not run flat-out; fix the pool URL, credentials or link and it resumes.
- “temperature too high” / protection → hydraulic fault; restore flow and coolant temp.
- PSU / power errors → failing or under-fed supply; verify the 200–240 V feed, then the PSU.
- “no hashboard” / missing board → re-seat the signal ribbon and bus-bar bolt; if still absent, the board or its domain has failed.
- Chip-test / EEPROM abnormal → a board failed self-test; note the board and chip index for repair.
Common mistakes
- Chasing “fan” faults — hydro units have none; the cooling fault is on the water side.
- Assuming a running pump means flow — a closed valve or air lock starves a board while the pump spins.
- Running a hydro miner on 110/120 V — the APW12-class PSU is 200–240 V only and will not make rated power.
- Opening the PSU right after shutdown — the primary caps still hold ~410–420 V. Wait and verify.
- Condemning a board for a temperature or zero-hash reading that is really a starved coolant path.
When to escalate
If a clean reset will not restore an IP, a board stays at 0 GH/s after confirmed coolant flow and a re-seated ribbon/bus-bar, or the kernel log flags chip-test/EEPROM failure, the unit needs board-level work — domain rework, chip repair or a control-board swap, not a field fix.
Related: Cross-reference symptoms with the ASIC fault finder, source cold-plate O-rings, fittings and replacement hashboards from ASIC repair parts, and when a board needs bench work, start a repair.
