New ANTMINER and PSU: Do’s and Don’ts

A new Antminer is two fragile machines bolted together: a power supply carrying 410–420 V DC on its PFC capacitors and pushing 130–270 A out the low side depending on model, and hash boards whose ASIC domains run on rails as low as 0.32 V. Most “dead on arrival” units that reach our bench were killed in the first ten minutes: a hot-plugged signal ribbon, a high-voltage-only PSU dropped onto a 120 V leg, or an APW that never matched the miner. A hash board is four figures. These checks are free.

What the PSU delivers, and to what

An APW12 is two rails with different jobs. OUT1 (the heavy J3/J4 connectors) is the adjustable high-current rail feeding the hash boards: 12–15 V, up to 233 A, 3,600 W at 220 V, set over I²C by the control board. OUT2 (J6) is a fixed 12 V rail, 15 A max, for the control board and fans. The APW17 uses the same two-rail split, rated 267 A on OUT1. That split is why a miner can boot, serve its web UI and spin fans with the hash boards unpowered.

On the board, that 12–15 V hits a MOS switch, then a boost converter raising it to 19–25 V, and only then do per-domain LDOs step it back down. Voltage is regulated per domain — a group of chips sharing one regulator — never per chip. An S19 runs 76 BM1398 chips in 38 domains (2/domain, ~0.36 V); an S19 Pro packs 114 BM1398 chips into the same 38 domains (3/domain, ~0.32 V); an S21 runs 108 BM1368 chips in 12 domains at ~1.2 V.

Match the PSU to the miner before you plug anything in

The 120 V question, settled

• APW3, APW5, APW7, APW8, APW9 and APW9+ do run on 110/120 V at reduced output, whatever the label implies. An APW7 gives 1,000 W on 120 V instead of 1,800 W; an APW3++ gives 1,200 W instead of 1,600 W.
• APW12 (200–240 V) and APW17 (220–277 V) are high-voltage input only. No jumper, no setting. On the APW12, undervoltage protection trips below 180 V AC and kills the output.

“PSU bypass mode” does not change that. It is a firmware feature that skips the PSU model-string check at boot. It does not disable thermal or overcurrent protection, and does not make a 220 V-only PSU work on 120 V. Its legitimate use is the reverse: running a 19-series board on an APW3++ at 120 V (1,200 W derated, roughly 35–56 TH/s depending on model), accepting the much lower ceiling.

Receiving inspection, before anything is energized

1. Tip the miner and listen. Loose heatsinks rattle. A heatsink separated from a BM1398 or BM1368 die is a warranty claim today, not a thermal fault later.
2. Inspect the DC connectors and hash board lugs for browning, melted plastic or spread pins — discolouration means a prior high-resistance joint at 200+ A.
3. Spin every fan by hand. A grinding fan or a dead spot fails under load and takes a hash board with it.

Cabling: the order is not a suggestion

1. Negative (GND) power cable first.
2. Positive power cable second.
3. Signal cable (the 18-pin data ribbon) last.

Removal is the exact reverse: ribbon first, then positive, then negative. Never insert or remove the ribbon with power applied. Those are 3.3 V I²C and UART lines to the control board; hot-plugging them backfeeds the 3.3 V logic rail and kills the level shifters and the ASIC chain’s UART/I²C I/O.

One hash board is fed by one PSU. In a dual-PSU build you partition by board: PSU 1 takes boards 1 and 2, PSU 2 takes board 3. Never split one board’s input across two supplies. Torque the DC lugs and re-check them after 24 hours of thermal cycling — at 200+ A a loose lug is a heater, and that is how connectors melt.

Power-on: what the clock should look like

• 0–2 s — PFC stages reach 410–420 V DC, PSU fan spins, auxiliary 12 V (OUT2) comes up and stabilises, control board boots.
• 10–30 s — Linux up, mining software starting, I²C bus available.
• 30–45 s — firmware asserts EN, reads the PSU version string, sets target voltage over I²C, verifies it back with the PSU’s ADC, arms the watchdog.
• 45–90 s — ASIC enumeration, PLL ramp, per-domain voltage optimisation, first shares to the pool.

No shares at two minutes means stop and diagnose, not power-cycle. Boards that go dark about a minute after a hang are the PSU watchdog working: the control board re-sends the watchdog every 10–30 s, and the PSU drops OUT1 (not OUT2) on a ~60 s timeout — the fans and web UI stay alive while the boards go dark.

Circuit and environment

• An APW12 at full 3,600 W output, at 93–95% efficiency, draws 3,790–3,870 W from the wall. A 240 V/20 A branch allows 3,840 W continuous under the 80% rule — too tight. Put a full-power 19-series on 240 V/30 A (5,760 W continuous).
• On 120 V, a 15 A circuit gives 1,440 W continuous and 20 A gives 1,920 W. An APW3++ at its 1,200 W ceiling pulls ~1,290 W in: a 15 A breaker with nothing else on it.
• PSU ambient 0–45 °C, 47–63 Hz AC. Keep the PSU intake out of the miner’s exhaust. No misting or water curtains near hash boards — condensation and airborne salt attack the PCB.

Where the actual hazard is

The mains side is the dangerous part: the PFC capacitors sit at 410–420 V DC and stay charged after you unplug the cord. Do not open a PSU. The hash board’s DC domains (0.32–1.2 V) and its 12–15 V input are not a shock hazard — but with hundreds of amps behind them they are an arc and burn hazard, so take the watch off. And never bench a hash board by shorting EN to GND: with no I²C master present the PSU comes up at its maximum default voltage, well above the board’s setpoint.

Common mistakes that kill new hardware

• Reseating the signal ribbon on a live miner.
• Buying an APW12 1215 for an L7 (it needs the 1417 window), or an APW17 for an S19.
• Assuming a 120 V outlet plus “bypass mode” will run an S19 on its stock APW12. It will not.

Related

• Testing a PSU with a multimeter
• Power on/off guidance for the 19-series PSU
• PSU problems and solutions: APW3 / APW3++ / APW5 / APW7
• ASIC repair parts — PSUs, fans, cables, control boards. If commissioning went wrong, start a repair.