How to fix “lacks hashboard” on Bitmain Antminer L7 ASICs

On an Antminer L7, lacks hashboard means the control board finished its chain scan and one of the four hashboards enumerated zero BM1489 chips. Each L7 chain carries 120 BM1489 ASICs and is worth roughly 2,376 MH/s of the machine’s ~9,500 MH/s nameplate (120 chips × 19.8 MH/s at stock ~3,425 W) — a full quarter of the machine’s output. The error clears only when all four chains enumerate their full chip count. The fault is almost never “the board died” — it is one broken link in a 120-chip daisy chain.

What the error actually reports

The control board addresses each hashboard as a UART daisy chain over the ribbon cable, 115200 baud by default, every Bitmain frame opening on the preamble 0x55 0xAA. Commands run forward (CI→CO, chip 1 to chip 120) alongside CLK, RST and BI/BO; nonces come back (RO→RI). Any chip that fails to forward CI, CLK or RST makes every chip behind it invisible.

A partial count — 60 of 120 — is the friendly case: the break sits at the chip immediately after the last one detected. A chain that enumerates zero chips is what prints lacks hashboard, and it means the failure is at the front of the chain, or the board has no rails, no clock, or no comms path at all.

The L7-specific detail that decides your first move: the L7 hashboard runs two 25 MHz crystals, not one — Y1 clocks chips 1–60, Y2 clocks chips 61–120. A dead Y2 gives you 60 of 120 and a low-hashrate complaint. A dead Y1 gives you zero chips and this error. If you are reading lacks hashboard, Y2 is not your problem.

Likely causes, ranked

1. Ribbon cable or connector. Cheapest and most common. It carries the 3.3 V logic rail, RST, the I²C pair for temp sensors and EEPROM, and the CI/RO UART pair. A bent pin kills enumeration with a healthy board behind it.
2. Voltage domain fault. BM1489 chips sit in voltage domains — blocks of chips sharing one regulated rail, wired in series so V_domain = V_per-chip × chips-per-domain. Voltage is regulated per domain, never per chip. A shorted chip or a shorted domain LDO drags its domain toward zero and the boost stage current-limits; an open LDO or an open chip leaves that domain high or absent. Either way the chain breaks and the board reports zero.
3. Boost / power-enable stage dead. The PSU rail reaches the domain regulators through a MOS switch and a boost converter. Dead MOS or dead boost means no domain comes up and the board sits cold with zero chips.
4. Crystal Y1 dead. Cracked crystal or a cold joint on its 100 nF loading cap. No clock at the head of the chain, no enumeration.
5. EEPROM fault or mismatch. The EEPROM holds board ID and calibration data, and all four boards must present coherent data. Corruption, a stuck I²C bus, or a donor board with foreign EEPROM data all read as “not there”.
6. Front-of-chain chip failure. A dead or shorted chip near chip 1 — or one that lives but stops forwarding CO/CLKO/NRSTO through a cracked BGA joint.

Diagnose before you desolder anything

Four identical chains is a free swap matrix. Change one variable per power cycle.

1. Read the kernel log, not the dashboard. The status page says a board is missing; the log says which chain index reported zero.
2. Swap the ribbon cable with a known-good chain. If the fault follows the cable, you are done for the price of a cable.
3. Swap the board into a known-good slot. If the fault stays with the slot, the problem is the control board port, not the hashboard.
4. Swap the PSU output pair. Stock L7 draw is ~3,425 W against the APW12’s 3,600 W rating — thin headroom, and a degrading PSU leg can drop a chain under load.
5. Reflash and re-scan. A corrupt image can misreport chain count. Only when the fault survives every swap does the board go to the bench.

Electrical measurements

Safety. The low-voltage DC domains on an L7 hashboard are a short-circuit hazard, not a shock hazard. The PSU is the dangerous part: the APW12’s mains side can hold roughly 400 V on its PFC capacitors after you unplug it. When probing, keep the black probe off the heatsink — touching the radiator shorts the board.

1. Unpowered, 30 seconds after disconnect. Meter in resistance mode, black probe on board ground, red probe on each domain test point. Chase the outlier: markedly low impedance = short, markedly high = open.
2. Power input resistance across the main connector. Near-zero means a short on the power bus — MOS, boost, or a failed filter cap. Never power a board that reads near-zero.
3. Powered, on a fixture. Wait 2–5 minutes for rails to stabilise, then sweep every domain on DC millivolts. All domains within ±50 mV of each other is healthy. One domain 100 mV or more below average is a partial short; significantly above average is an open. Reference a known-good L7 board.
4. Clock and signal integrity. Scope CLK at Y1 and between the first two chips. BM-series typical bands: CLK 0.58–1.3 V, CI/CO 0–1.8 V, RI/RO 0.3–1.8 V, RST 0–1.8 V. If CLK is absent at chip 1, stop. You have found it.

Board-level repair

• Inspect at 10x first. The series resistor on the CLK line at each domain boundary is a documented cold-joint site — probe CLK either side of it. So are the CI/BO series resistors at each chip, the 100 nF capacitor that couples CLKO into the next domain’s CLKI, and the crystals.
• Reflow before you replace. A cracked BGA joint under a chip that still lives will re-seat with a proper profile. Use a real reflow station, not a heat gun.
• Replace only what you have proven dead. If the domain sweep and the signal probe both point at one BM1489, replace that chip. Shotgunning chips is how a repairable board becomes scrap.

Verify the fix

1. Fixture-test the board alone and confirm 120 of 120 chips — not 118, not 119. A short chain passes a casual glance, then fails in the field.
2. Reinstall, confirm all four chains report full counts, then run it an hour watching per-chain hashrate and temperature. A board that enumerates cold and drops out warm is a thermal joint failure, not a fixed board.
3. Confirm it settles near profile — ~9,500 MH/s at ~3,425 W stock. A full chip count at a quarter-low hashrate means you fixed enumeration and left a fault behind.

Common mistakes

• Chasing Y2 on a zero-chip board. Y2 only clocks chips 61–120. If Y2 fails you get 60 chips, not lacks hashboard.
• Trying to run an L7 PSU on 120 V. The APW12 (14–17 V variant, the one the L7 uses) is 200–240 V input only. Older APW3/APW5/APW7/APW8/APW9 units do run on 110/120 V at reduced output despite their labels — the APW12 does not.
• Reflashing forever. Zero chips after a clean flash and a cable swap is hardware. Firmware will not fix a dead crystal.

Related

• Removing the hashboard from an ASIC miner
• ASIC Fault Finder
• ASIC repair parts
• Start a repair