← All guides

ASIC PSU failure signs: the four signatures and the repair-or-replace call

Hardware · 8 min read

Educational content only, not financial advice. Figures below come from the published sources named in the text and were checked on 1 November 2026; hardware prices and network conditions change, so re-check before spending anything.

Four. That is how many failure signatures a repair guide for the APW9 and APW12 families uses to sort a dead Antminer before any money moves: no fan and no output, fan spinning but no DC output, output that stops and restarts in a cycle, and normal output with a fan that will not turn. Each signature rules out a whole class of cause. This guide verifies the specifications that decide which supply you even have, walks the signatures in order as a procedure, and ends with the repair-versus-replace call, including a dated Canadian price for a replacement, checked 1 November 2026.

First, confirm which supply is on the bench

APW12 is a family name, not a single part. Bitmain's own support note lists three main models. APW121215, APW121417 and APW12A, and states plainly that they are not interchangeable. The four digits encode the output window: 1215 means 12 to 15V, 1417 means 14 to 17V. Within the 1215 line, versions a, b and c are interoperable, and versions d, e and f can be used with each other; a d/e/f unit can replace an a/b/c unit after a firmware upgrade, but not the other way round. Bitmain's warning for a mismatched swap is blunt: the miner may be unusable, or the supply may burn out. Read the label before ordering anything.

SupplyOutputPairs withInput note
APW914.5 to 21V, 170A, about 3,600W at 220VS17, S17 Pro, T17Starts on 110/120V at sharply reduced output, expected, not a fault
APW9+14.5 to 21V, 200AS17+, S17e, T17+, T17eSame reduced-output behaviour on 120V
APW12 (1215)12 to 15V, 233A, about 3,600WS19, S19 Pro, S19j, S19j Pro, T19200 to 240V only; undervoltage protection trips below roughly 180V
APW12 (1417)14 to 17V variantL7, K7, KA3 and similarNot a substitute for a 1215 unit

That input fork explains a classic false alarm. An APW9 fed 120V starts and runs weakly. An APW12 fed 120V refuses to start at all, because its undervoltage protection is doing its job. An S19 that will not power up on a 120V circuit is a wiring decision, not a broken power supply.

Safety: the DC side is not the dangerous side

The output rails sit at 12 to 21V and are not a shock hazard. The hazard is inside the case on the AC side: the power-factor-correction bulk capacitors hold roughly 410 to 420V and stay charged for minutes after the cord is pulled. Opening a supply means discharging them through a proper bleeder resistor and confirming 0V before touching the primary side. Never hot-plug the 4-pin signal connector either; its 3.3V logic is easy to kill. Burned smell, a melted connector, corrosion or liquid exposure ends the home diagnostic on the spot, those go to a repair bench.

Short version: diagnose outside the case. Repair inside it is a bench job.

The four signatures, worked in order

1. No fan, no output. Treat this as an AC-side problem until proven otherwise. Confirm the cord is fully seated at both ends and rated for the load, a 3,600W unit needs a heavy-gauge C19/C20 cord on a dedicated circuit, not a thin kettle lead. Meter the outlet under load, check the breaker or PDU port, then swap in a known-good cord and outlet. If AC is confirmed good at the inlet and the unit stays lifeless, the fault is internal, a blown input fuse or a failed PFC stage, and that is bench repair, not a field fix.

2. Fan spins, no DC output. The primary side is alive; the unit is starved or sitting in protection. On an APW12, meter the AC input and confirm a solid 200V or more under load, a sagging shared 208V circuit can dip below the start threshold. Check the output side for a short or overload: a shorted hashboard or a pinched bus-bar lug drops the supply straight into short-circuit or overpower protection. If it tripped, power off for a full 10 minutes so the protection latch and thermal mass reset, then re-apply power with the load disconnected. Clean output with no load means the fault is downstream in the miner, not in the supply.

3. Output stops for a few seconds, then resumes, repeatedly. This hiccup is the classic over-temperature signature: the unit shuts down, cools slightly, recovers, heats, and trips again. Watch the internal fan, clear blocked ducts and filters, blow out dust-packed heatsinks, and check that load and ambient temperature sit inside the unit's limits. One caveat worth knowing: with a healthy fan and clear airflow, a cyclic dropout can also be a control-board watchdog timeout, a crashing control board stops sending its heartbeat, and the supply drops the hashboard rail after roughly 60 seconds as a safety feature. That points at the control board, not the supply.

4. Output normal, fan not spinning. The power stage is fine and the cooling fan has failed. Clear any debris jamming the blades. If the blades are free and the fan still will not turn, the fan or its internal 12V rail has failed. Do not keep mining on it: signature 4 left alone becomes signature 3, then a dead supply.

What success looks like: after the fix, run the miner long enough to confirm stable accepted hashrate, normal fan behaviour, steady chip temperatures and a normal reject rate on the pool side. A dashboard that looks fine for five minutes proves nothing about a recurring power fault.

Protection trips are information, not obstacles

The APW9 and APW12 hardware protections run independently of firmware and cannot be disabled. Overvoltage needs a full power cycle to clear. Undervoltage clears when proper voltage is restored. Overcurrent, at roughly 250A, current-limits or shuts down and usually auto-recovers. Over-temperature auto-recovers once cool. Short-circuit needs the short removed, then a power cycle. Overpower, above roughly 3,600W, current-limits until the load drops. A supply that keeps tripping under a normal load is reporting a real fault. Find it; do not defeat the protection.

Repair, bench repair, or replace?

Three findings point to an internal fault: no output with confirmed-good AC, a dead internal fan, or a unit that trips protection under a known-good load. Those are bench-level repairs with real parts. Before spending either way, isolate the supply from the miner and bench-test it against a known-good load, that single test separates a bad supply from a bad machine.

The replacement side of the ledger, checked 1 November 2026: a refurbished, bench-tested APW12 1215 from D-Central, shipping from Montreal, was listed at C$249.99 and in stock, with a 30-day window to report a defect or dead-on-arrival unit. Against that price, a bench diagnosis makes sense when the miner itself is worth keeping and the fault is unconfirmed; a straight replacement makes sense when the bench test has already condemned the supply and downtime costs more than the part. Either way, match the full model and version from the label, 1215 or 1417, and the version letter, before the order goes in.

, Dex Carter, Tech Writer