How to fix “error power lost” on Bitmain Antminer T17 ASICs

On an Antminer T17, ERROR_POWER_LOST does not mean the miner lost mains power. It means bmminer refused to hash: the DC rail it commanded from the APW9 is not the rail it measured back. The T17’s PSU is a variable supply (14.5–21 V DC, 170 A, 3600 W at 220 V): the control board sets its output to the string voltage the hashboards need, then reads the value back. Diverge by more than 1.0 V and the firmware aborts rather than run 12 voltage domains at an unverified voltage. The result is a miner sitting at zero hashrate with no chains initialised, while the real fault — usually the AC side or a DC connector — goes unaddressed.

What the error actually reports

Pull the kernel log (see how to copy the kernel log into text form). You will see three lines:

• check voltage: target_vol = 16.90, actual_vol = 18.21, more than 1.0v diff — commanded PSU setpoint vs. measured output.
• ERROR POWER LOST — the flag raised by that comparison.
• stop mining: power set failed — the abort. No chains initialise after this point.

Two things follow. First, those are PSU rail volts, not chip volts. A 17-series hashboard carries 12 voltage domains in series — voltage is regulated per domain, never per chip — and their sum is what the APW9 must hold, which is why a healthy target sits in the mid-to-high teens. Second, the direction of the mismatch tells you where to look. Actual below target: the rail is being pulled down — weak AC feed, high-resistance connection, dying PSU, or a shorted board loading it. Actual above target (18.21 V against a 16.90 V setpoint): the rail is running higher than commanded. Two causes — the control board never wrote the setpoint or never read it back (the PSU communication path), or a domain or board has gone open-circuit and the string is drawing less current than the supply expects. Check the PSU comm path first, then sweep the domains for an open.

Likely causes, most common first

1. AC supply that sags under load. Documented T17 operating points span 1,570 W (600 MHz / 36 TH) to 2,910 W (900 MHz / 54 TH). The APW9 needs 200–240 V for its 3600 W rating; it will run on 110/120 V, but at sharply reduced output, and a 2 kW+ load on a 120 V circuit is exactly what collapses the rail mid-ramp.
2. DC output cabling and lugs. At well over 100 A, a few milliohms of contact resistance is volts of drop and a lot of heat. Loose screws, oxidised lugs and heat-browned insulation at the board end are our most common bench finding here.
3. A degraded APW9 that cannot hold its setpoint under load. Reads fine unloaded, droops the moment three boards ramp.
4. A shorted domain, chip or filter capacitor on one hashboard. The short pulls its domain toward zero, the supply current-limits, and the measured rail lands under target: one bad board takes the whole miner down.
5. The PSU communication path. If the control board cannot write the setpoint or read the voltage back, the comparison runs on garbage: power set failed with no electrical fault present.
6. Firmware or setpoint corruption after a failed or mismatched upgrade. Last to suspect, never first.

Diagnose before you desolder anything

• Move the miner to a known-good 200–240 V circuit on its own breaker, shortest cord you have. If the error clears, there was never a hardware fault.
• Re-terminate the DC cables. AC unplugged, back out every PSU-side and board-side screw, inspect for oxidation and heat damage, clean, re-torque.
• Reseat the PSU communication cable at both ends with AC removed. Never hot-plug it — the logic on either end can be damaged by a live insertion.
• Isolate by chain. Power removed, pull one hashboard’s feed (signal cable first, then positive, then ground; reassemble in reverse) and boot. Ignore the missing-chain complaint; watch whether check voltage now passes. Rail holds with board N out, collapses with it in → board N is the fault.
• Swap in a known-good APW9 — but only after the above.

Electrical measurements

Safety, stated correctly: the low-voltage DC side of a T17 is not a shock hazard — but a 170 A rail will vaporise a probe tip and weld a dropped screwdriver. The PSU is the dangerous part: its PFC bulk capacitors sit at 410–420 V DC and stay charged after you unplug it.

1. AC at the PSU inlet, under load — not at an idle outlet. Sag out of the 200–240 V band during ramp is your answer.
2. DC at the PSU output terminals, under load. Compare against target_vol from the log. The firmware aborts past 1.0 V of divergence. A healthy supply tracks its setpoint closely — a rail that drifts a few tenths of a volt under load, even without tripping the 1.0 V abort, is a supply on its way out.
3. DC at the hashboard lugs, under load, same instant. Any real difference between PSU terminals and board lugs is being dropped in the cable or the joint — that delta is the fault.
4. Board input resistance, unpowered. Wait 30 s for capacitors to discharge, then measure across each board’s DC input. Near-zero is a short on the power bus (MOSFET, regulator, failed capacitor): that board is what drags the rail down.
5. Domain sweep. Black probe on a board ground point, never the heatsink. Walk the 12 domain test points: consistent impedance across all twelve is healthy, one far lower is a short, far higher is an open.

Our PSU multimeter guide covers load-testing; APW9/APW12 problems and solutions covers this supply’s failure modes.

Board-level repair

• Find the short by domain, not by chip. The impedance sweep names the collapsed domain. Inject a low, current-limited voltage across it from a bench supply and thermal-image the board: the failed part heats first. A shorted filter capacitor is the most common find; a shorted hashing chip is next.
• Replace, do not just remove. Lifting a shorted capacitor restores the domain, but a shorted chip must be replaced: pulled and left out, it breaks the daisy-chained signal path and turns this error into a chip-count error.
• Check the board’s power input for a shorted MOSFET before blaming the domain chain.

Verify the fix

1. Reassemble ground cable, positive cable, signal cable last; boot on 200–240 V.
2. Boot it at its lowest power setting first — the lowest documented operating point (~1,850 W / 40 TH) — before pushing it toward its 2.9 kW ceiling.
3. Re-read the kernel log: check voltage lines where target and actual track, no power set failed, all three chains enumerating.
4. Hash it two hours at temperature. Contact-resistance faults return as the joint heats: ten clean minutes is not a pass.

Common mistakes

• Replacing the PSU first. Cables, lugs and AC feed beat PSU failure by a mile.
• Running a T17 on a 120 V circuit and blaming the miner. The APW9 will start there; it will not hold 2 kW+ there.
• Raising the frequency to push through the error, which only raises the current the sagging rail must deliver.

Related

• Hashboard voltage is inconsistent — the sibling fault: domains are off, rail still holds.
• ASIC Fault Finder — match your exact log string to a repair path.
• Start a repair — our bench does 17-series domain-level rework, including chip replacement.