How to fix “hashrate is low or a hashboard is faulty” on Bitmain Antminer L3+ ASICs

On an Antminer L3+ this is a fault category, not a literal kernel-log line: cgminer either sees fewer chips than expected on one of the four chains, or sees all of them and still misses work. The L3+ carries four hashboards of 72 BM1485 chips each — 288 chips total — and the 504 MH/s nameplate at the stock bitmain-freq:384 works out to 126 MH/s per board, ~1.75 MH/s per chip. One chain that fails to enumerate drops you to ~378 MH/s while the fans keep burning; a chain that enumerates but errors out pulls full domain current and returns nothing.

What the error actually reports

Stop reading the dashboard; pull the numbers: echo '{"command":"stats"}' | nc <miner-ip> 4028. Four fields decide everything.

• chain_acn1–chain_acn4 — chips detected per chain. A healthy L3+ reads 72 on all four. Less is a chain break.
• chain_acs1–chain_acs4 — per-chip status, o responding, x silent. BM1485 chips daisy-chain CI/CO, CLK and RST across the board, so every chip downstream of a dead one goes silent too. A long tail of x is not 40 dead chips — the break is at the first x.
• chain_hw1–chain_hw4 — hardware errors. 72/72 detected plus a climbing chain_hw means the chips are alive but returning bad nonces. Heat, a sagging domain, or too much frequency — not a dead chip.
• chain_rate1–chain_rate4 — per-chain MH/s. Expect ~126 each.

Likely causes, ranked

1. Heat and dead thermal contact. L3+ heatsinks rely on the thermal interface, not on a solder bond. Dried-out paste, a lifted heatsink, or a lint-packed fin stack drops chips intermittently: enumerates on a cold boot, degrades twenty minutes in.
2. Ribbon cable and connector. The signal ribbon carries 3.3 V logic, RST, the UART pair (CI/RO) and the temp-sensor I²C — but not the hash clock, which is generated on the hashboard itself. A cracked conductor kills enumeration and temp reads, not the clock.
3. PSU sag. A stock L3+ pulls about 800 W at the wall. An APW3++ does run on 110/120 V, derated to 1200 W; an APW7 on 110/120 V gives only 1000 W (1800 W at 220 V) — ~80% of that ceiling. Sag shows up as one chain shedding chips under load.
4. A failed voltage domain. Each board is 12 domains × 6 chips — ~0.80 V per domain at 270 MHz, ~0.83 V on the stock 384 MHz tune. A shorted chip drags its domain toward zero; an open one lets it rise.
5. Chain break at a chip. Cold solder on a chip’s CO/CLKO/NRSTO pins, or on the CLK/CI/BO resistors at a domain boundary, kills signal forwarding while the chip still lives.
6. Frequency margin gone. 2017-era parts with hours on them: a board once clean at 384 MHz may not be.

Diagnose before you desolder anything

All free or near-free. Re-read chain_acn after each step.

1. Blow the boards out and check every heatsink on the suspect chain is still tight against its chip.
2. Swap the ribbon cable with a known-good chain’s. If the fault follows the cable, you are done.
3. Swap the PCIe power cables with those feeding a good board; feed every connector on a board from one PSU.
4. Move the board to a different controller port. Fault follows the board → the board is bad. Fault stays on the port → the control board is bad.
5. Boot with only the suspect board attached and read the chain_acn field for the port it is plugged into (port 3 → chain_acn3). That ends any argument about a shared rail or bus.
6. Reset the tune. Back up first: cp /config/cgminer.conf /config/cgminer.conf.bak. /etc/cgminer.conf.factory holds the stock bitmain-freq:384 but also factory pool entries, so copy just the bitmain-freq value back to 384, or restore the file and re-enter your pools. Used L3+ units arrive carrying someone else’s overclock; a board that fails at 450 MHz is not faulty.

Electrical measurements

Safety, stated correctly: the hashboard is not a shock hazard — ~9.6–10.0 V across the board, ~0.80 V per domain at 270 MHz and ~0.83 V on the stock 384 MHz tune; what matters is the domain-to-domain spread (±50 mV), not the absolute number. The PSU is: its mains side holds ~400 V on the PFC caps after you unplug it. Probe the board freely; never the PSU.

1. Unpowered, resistance mode. Black probe on a board ground pad — never a heatsink, which shorts across a domain. Walk the 12 domain test points; impedance should be consistent. Markedly low = a short.
2. Power-input resistance across the board’s 6-pin inputs should be several ohms. Near zero means a shorted MOSFET, boost stage or input cap — do not power that board up.
3. Powered, DC volts. Stabilise 2–5 minutes, then read all 12 domains. All within ±50 mV of each other = healthy. Over 100 mV below average = a partial short (failed chip or dead cap). Well above average = an open.
4. Total rail. The 12 domains should sum to ~10.0 V on the stock tune (~9.6 V at 270 MHz). A low rail with no bad domain = PSU or cable fault.

Board-level repair

• Locate the break by dichotomy. Short the RO test point to the domain rail at the midpoint of the chain and re-enumerate. More chips → the fault is after the midpoint; same or fewer → before it. Halve again — about seven passes.
• Inspect before you heat. At 10×, check the suspect chip’s CO/CLKO/NRSTO joints and the domain-boundary CLK/CI/BO resistors. A reflow fixes a cold joint; nothing fixes a chip you cooked chasing one.
• Shorted chip: it comes off. Replace, reball, then re-measure that domain before re-enumerating — if it is still low, the regulator or a cap went with the chip. One dead chip is an economic repair; several collapsed domains are not.

Verify the fix

• chain_acn = 72 on all four chains, every chain_acs string all o.
• chain_hw flat — not necessarily zero, but not climbing — over two hours.
• chain_rate ~126 MH/s per chain, ~504 MH/s total — read the long-window average, not the 5-second figure.
• Enumerates 72/72 but chain_hw still climbs? It needs less frequency. The BM1485 rows: 270 MHz → ~9.6 V/~480 W/~360 MH/s; 320 MHz → ~9.8 V/~600 W/~432 MH/s (the usual landing); 384 MHz → 10.0 V/800 W/504 MH/s (nameplate). bitmain-freq is global on stock firmware — slowing one repaired board slows all four.

Common mistakes

• Hunting for per-chip voltage control. There is none. The BM1485 rail is regulated per domain, six chips to a domain — one twelfth of a board is your finest granularity.
• Blaming the board for the PSU. Prove the rail first.

Related

• ASIC Fault Finder — symptom to subsystem.
• ANTMINER ASIC status showing ‘X’
• “crc check error” on the L3+ — the sibling ribbon/UART fault.
• Testing a PSU with a multimeter
• Start a repair with D-Central — our bench does chip-level BM1485 work.