ASIC PSU Repair Guide: APW3, APW7, APW9, APW12 Failure Modes, Safety & Diagnosis

APW PSU Model Quick Reference

Bitmain produces a distinct PSU for each generation of miner. Model selection matters — voltage ranges, connector types, and control protocols are not interchangeable. The following table is sourced from the Bitmain APW specification documents and the open-source APW12 protocol documentation (skot/bitcrane_test_scripts, 2026-03-09).

Note: The APW3++ accepts 100–264 V AC (universal input), making it the standard choice for S19-series 120 V home mining via PSU bypass mode. The APW9/APW12/APW17 require a 200–240 V supply and are not 120 V compatible.

APW12 Internal Architecture

The APW12 (used in S19-series miners) has a dual-output design that is worth understanding for accurate fault diagnosis:

• OUT1 (J3/J4): The main adjustable rail — 12–15 V, up to 233 A (3,600 W max) — feeds the hash boards.
• OUT2 (J6): A fixed 12 V, 15 A rail that powers the control board and cooling fans. This stays alive even when OUT1 is tripped by a protection event.
• PFC stage: Two parallel PFC stages charge large bulk capacitors (C1/C2) to 410–420 V DC. This is the high-voltage bus referenced in the safety warning above.
• LLC conversion: Four MOSFETs (Q14, Q15, Q31, Q32) perform the main DC-DC conversion from the 410 V bus down to the output rail.
• PIC16F1704 MCU: Manages I2C communication, watchdog timer, and voltage DAC. This is the chip that fails on versions a/b/c when used where d/e/f/g is expected.

Common Failure Modes by PSU Model

APW3 / APW3++ (S9 era)

• Fan bearing failure (most common): The single internal cooling fan degrades and eventually stops. Thermal runaway follows. Symptom: grinding noise before total silence, then the miner shows no hash rate.
• Bulk capacitor degradation: Electrolytic capacitors dry out over multi-year operation, increasing ripple on the 12 V rail. Symptom: voltage instability under load, intermittent hash board faults.
• PFC diode bridge failure: Heavy-load environments stress the input rectifier. Symptom: tripped breaker, blown input fuse, no output at all.
• Output connector arcing: The 6-pin PCIe connectors oxidize or loosen under high current. Symptom: localized heat, discolored plastic, voltage drop on affected hash board.

APW7 (S9 / L3+ era)

• Same fan and capacitor failure modes as APW3.
• Higher current path damage: APW7 delivers 150 A maximum vs. APW3’s 133 A. At sustained high loads, PCB traces and output connectors see more stress. Connector arcing is more prevalent on degraded APW7 units than on APW3.
• Input bridge rectifier failure: Same failure mode as APW3 PFC diode, appears more frequently in hot or dusty environments.

APW9 / APW9+ (S17 era)

• OTP shutdown at high ambient temperature: The S17 family runs hot and the APW9’s ~90% efficiency (lower than APW12) generates more internal heat. In summer or poorly ventilated facilities, the OTP protection trips repeatedly. Symptom: miner drops offline during hot periods, recovers when cooled.
• Fan failure: Same mode as other APW units. S17 hashcenters see this frequently due to dust accumulation.
• PFC MOSFET failure: At 3,600 W input, the APW9 PFC MOSFETs are more stressed than lower-wattage predecessors. Failure is typically a hard short — loud pop, blown fuse, no recovery.
• Voltage output instability: The APW9’s higher output voltage range (14.5–21 V) means a drifting DAC or reference can cause hash boards to receive out-of-spec voltage, triggering board faults that look like board failures.

APW12 (S19 era — most common in active fleets)

• Fan failure: The APW12’s internal fan is not I2C-controllable (it is managed by internal thermal circuitry only). When it fails, OTP trips. Symptom: miner goes offline, LED indicators may remain lit (OUT2 is still active), but hashrate drops to zero.
• APW121215a/b/c vs. d/e/f/g version mismatch: Versions a/b/c lack voltage feedback capability. If an a/b/c unit replaces a d/e/f/g unit, the miner firmware cannot close its voltage regulation loop — it will either refuse to start or run incorrectly. You can replace a/b/c with d/e/f/g (with a firmware update), but not the reverse.
• I2C communication failure: The PIC16F1704 MCU manages the I2C bus (address 0x10, bus speed 400 Hz — note: 400 Hz, not 400 kHz). If this chip fails or its firmware is corrupted, the miner cannot set output voltage or verify PSU state. Symptom: miner reports PSU communication error at boot.
• LLC stage MOSFET failure: MOSFETs Q14, Q15, Q31, Q32 can fail under sustained overload or a hash board short-circuit event. Symptom: loud pop, SCP protection trips, no output voltage, may blow input fuse.
• OCP trip from hash board fault: A shorted or partially failed hash board can pull well over 250 A, triggering the OCP. The PSU then appears dead. Remove the faulty hash board first and retest before condemning the PSU.

APW17 (S21 era)

• Uses the same I2C frame format as APW12 (preamble 0x55 0xAA, slave address 0x10, write register 0x11) but reports firmware strings beginning with “APW17_” rather than “APW12_”. The physical power connector is different from APW12 — an adapter cable is required for cross-testing.
• Max current of 267 A (vs. APW12’s 233 A). Failure modes are expected to be similar; the fleet is younger so long-term wear data is limited.
• Requires 220–277 V AC input. Not compatible with standard 208 V US circuits without verification.

APW12 Hardware Protection Triggers

Understanding which protection tripped helps narrow the fault before disassembly. All APW12 hardware protections are firmware-independent — they cannot be disabled by the control board or by third-party firmware.

Note: The APW12 I2C protocol exposes only an ON/OFF state flag (command 0x05). There are no detailed fault registers accessible via software — unlike PMBus-compliant supplies. Fault identification requires hardware diagnostic methods described in the checklist below.

Pre-Diagnosis Checklist

Work through this checklist in order before disassembling or condemning a PSU. Many apparent PSU failures are caused by the load (hash boards, cables) rather than the PSU itself.

Step 1 — Isolate the PSU from the miner

• Power down and unplug the miner completely.
• Disconnect all hash board power cables from the PSU output (J3/J4 connectors).
• Disconnect the control board power cable (J6).
• Leave only the AC input connected.

Step 2 — Visual inspection (external)

• Inspect the AC power cord and inlet for burn marks, melting, or discoloration.
• Inspect the DC output connectors (J3/J4) for burn marks, bent pins, or melted plastic — connector arcing is a common damage mode on APW3/APW7 units.
• Inspect the PSU casing for deformation, burn holes, or evidence of a vent event.
• Plug in AC and listen for the internal fan spinning up within 2–3 seconds. Silence (when it was previously noisy or functional) often means fan failure.

Step 3 — Test output voltage without load

• With hash boards disconnected, power the PSU and measure DC voltage across the J3/J4 output connectors using a multimeter.
• APW12 (1215): expect 12–15 V DC (default ~13 V or maximum ~15.2 V if I2C is not setting voltage).
• APW3++: expect ~12.15 V DC (fixed output, no I2C control).
• No voltage with AC present and fan running usually indicates LLC stage or PFC stage failure.
• No fan, no voltage, no response: check the input fuse (if accessible) before further diagnosis.

Step 4 — Test the hash boards independently

• If the PSU shows correct no-load voltage, reconnect hash boards one at a time and observe whether voltage collapses.
• A hash board that immediately pulls voltage below 10 V is likely shorted and will cause OCP or SCP to trip — the hash board is the fault, not the PSU.

Step 5 — APW12/APW17 I2C verification (if miner firmware reports PSU error)

• On S19-family miners with third-party firmware (Braiins OS+, LuxOS), a “PSU communication error” at boot usually means the control board cannot reach the PSU on I2C bus address 0x10.
• Causes: damaged 4-pin J15 signal connector, failed PIC16F1704 on the PSU, or APW121215 version mismatch (a/b/c installed where d/e/f/g is required).
• The APW3 and APW7 do NOT use I2C — they are simple PSUs with no communication protocol. PSU bypass mode in third-party firmware is required if using APW3++ on an S19.

Step 6 — Consult the error codes

• On Bitmain stock firmware, PSU faults are logged in the miner’s system log. Check the Bitmain web UI under “Miner Status” or via SSH at /var/log/log.
• Cross-reference with D-Central’s open data hub, which includes the error code database covering APW-related miner fault codes.

APW12 Diagnostic Test Points (For Qualified Repair Technicians)

The following internal test points are documented in the Zeus Mining APW12 repair guide and the open-source community reverse-engineering of the APW12. All measurements require active discharge of the PFC bulk capacitors beforehand.

Source: Zeus Mining APW12 Repair Guide; skot/bitcrane_test_scripts (open source, CC BY). Test point locations vary slightly between APW121215 sub-versions — consult a version-specific schematic if available.

Repair vs. Replace Decision Guide

Not every failed PSU warrants repair. The calculus depends on parts availability, labor cost, the miner generation, and the failure type.

Repair makes economic sense when:

• Fan replacement only. An internal fan swap takes 15–30 minutes and the replacement fan costs under $20. A fan-failed APW12 with otherwise healthy internals is a strong repair candidate.
• External connector rework. Burned 6-pin PCIe connectors on an APW3/APW7 can be replaced with the correct Molex connector and proper crimp tool. Labor is 30–60 minutes; parts cost is minimal.
• The PSU is current-generation (APW12, APW17). Replacement cost is high enough that repair overhead is justified.
• Fault is clearly identified and isolated — a blown input fuse or a single shorted capacitor with an otherwise clean board is a well-defined, bounded repair.

Replace (or send for depot repair) when:

• PFC stage MOSFET failure. The MOSFETs and gate drivers in the PFC stage require specialized components, heat gun rework, and a load-test rig to verify repair. Depot repair or outright replacement is typically more economical unless you have in-house PSU repair expertise.
• LLC stage failure. Q14, Q15, Q31, Q32 failure with possible transformer damage. Complex repair with high risk of secondary damage.
• PIC16F1704 MCU failure on APW12. Replacing the MCU requires sourcing the correct chip, re-flashing with Bitmain PSU firmware, and re-calibrating the DAC. This is a depot-level repair.
• The PSU is APW3/APW7 vintage and supports a heavily depreciated miner (S9, etc.). Replacement cost may be lower than diagnostic labor for a PSU on a $50–200 miner. Use the ASIC repair cost estimator to model the numbers.
• Multiple failure modes present simultaneously. A PSU that shows fan failure, capacitor bulge, and a connector burn is not a single-fault repair — the internals have been thermally stressed multiple times.
• Damage from liquid infiltration or flood. Total replacement.

Not sure which way to go? Use D-Central’s Warranty vs. Repair decision tool — it walks through age, failure type, and replacement cost to give a clear recommendation.

D-Central PSU Repair and ASIC Repair Services

D-Central Technologies repairs ASIC miners and their power supplies from our facility. We have hands-on experience with APW3, APW7, APW9, APW12, and APW17 units across the S9, S17, and S19 miner generations.

• PSU fan replacement and cleaning — standard turnaround, most common service request.
• Connector rework and cable fabrication — burned 6-pin PCIe output connectors, J15 signal connector repair.
• Load testing and voltage calibration — post-repair verification at rated current, DAC voltage accuracy confirmation on APW12 units.
• Full ASIC repair — hash board rework, control board repair, firmware recovery. PSU diagnosis is included as part of full miner repair intake.

Submit a repair request at D-Central’s ASIC repair service page. Include your PSU model, miner generation, and the symptoms you observed — the more detail you provide, the faster we can triage.

Need to estimate the cost before committing? Use the ASIC repair cost estimator for a rough range based on failure type and miner model.

For context on the full S19-series ecosystem (which uses the APW12), see the Antminer S19 hub. For S21-series and APW17, see the S21 family comparison.

Additional Resources

• D-Central open data hub — includes the error code database, miner reliability index, and power profiles data (820 tuning profiles across 28 Antminer models)
• Power cost calculator — model electricity cost at various PSU efficiency points
• ASIC repair cost estimator — repair vs. replacement cost modeling
• Warranty or repair decision tool
• D-Central ASIC repair service

---

Frequently asked questions

Is the APW12 dangerous to open for repair?

Yes — the APW12’s PFC (Power Factor Correction) stage charges its bulk capacitors (C1 and C2) to 410–420 V DC. This voltage persists for several minutes after you unplug the AC supply. Before touching any internal component, wait at least 5 minutes, then use a multimeter to measure directly across those capacitors to confirm discharge. If you are not qualified to work safely on high-voltage switching power supplies, do not attempt internal APW12 repairs. Send the unit to a depot repair facility such as D-Central.

Can I use an APW3++ on my S19 for 120 V operation?

Yes, but with important limitations. The APW3++ accepts 100–264 V AC (universal input), unlike the APW12 which requires 200–240 V. When paired with an S19 on 120 V, you must enable PSU bypass mode in the miner firmware — this tells the firmware to skip the APW12 identification check and run without I2C voltage control. Power target should be capped at or below 1,200 W (the APW3++’s safe 120 V maximum), which will reduce hash rate to roughly 50–60% of the S19’s rated speed. All thermal protections on the miner remain active; the APW3’s own thermal protection remains active as well.

What is the difference between APW121215a/b/c and APW121215d/e/f/g?

The key difference is voltage feedback: versions d, e, f, and g include a voltage feedback loop that allows the miner firmware to read back the actual PSU output voltage via ADC (command 0x04) and close a regulation loop. Versions a, b, and c lack this circuit — they accept voltage setpoint commands but cannot confirm actual output. Versions d/e/f can replace a/b/c in the field (the miner firmware requires an update to take advantage of feedback, but it will still run without it). However, a/b/c units cannot replace d/e/f/g units in miners whose firmware expects feedback capability — the firmware will either refuse to start or will operate without accurate voltage regulation.

Why does my APW12 show no output voltage even though the fan is running?

The APW12 has two separate output rails. OUT2 (12 V fixed, 15 A) powers the control board and the internal fan — this rail stays active even when the main hash board rail (OUT1) is disabled or has tripped a protection. If you see the PSU fan running but measure zero voltage on the J3/J4 hash board connectors, the most likely causes are: (1) OUT1 tripped by OCP or SCP due to a shorted hash board — disconnect the hash boards and retest; (2) the EN pin signal from the control board is not asserting (control board fault, not PSU); (3) LLC stage failure — OUT1 cannot generate output even though the auxiliary rail is healthy. Work through the pre-diagnosis checklist above in order before opening the unit.

How do I identify which APW12 version I have?

With third-party firmware (LuxOS, Braiins OS+, or DCENT_OS), query the PSU firmware version string via I2C command 0x01. The response is a 16-byte ASCII string in the format APW12_1215x_Vn.m where “x” is the sub-version letter (a through g). On Bitmain stock firmware, check the Miner Status page — it logs the PSU firmware version at startup. Physically, the version is printed on the PSU label and may also appear on the PCB silkscreen near the PIC16F1704 MCU.

When should I replace my PSU rather than repair it?

Replace when: the PSU shows multiple simultaneous failures (fan, capacitor, and connector damage together — this indicates repeated thermal stress), when the LLC stage MOSFETs or PFC MOSFETs have failed (depot-level repair with specialized equipment), or when the failed PSU supports a low-value miner (S9-era hardware with market value under $100–200). Repair makes clear sense for isolated fan failures, connector rework, or fuse replacement on current-generation hardware (APW12, APW17). Use the repair cost estimator to run the numbers for your specific situation.