Bitaxe Ultra – BM1366 Hashrate Cliff / Sudden Drop

Hard power-cycle the miner. Unplug at the PSU for 30 seconds, not a soft reboot. Brings BM1366 and TPS546 back to a clean cold-boot state. If a soft state was wedged after an AxeOS update, this clears it. Quick, free, rules out the easiest cause.

AxeOS factory reset. Settings -> System -> Reset Settings. Confirm. Reboot. Observe 30 minutes of steady-state hashing. Healthy Ultra at stock: ~500 GH/s +/- 10% at ~15 W, ASIC < 70 C. If hashrate returns to nominal, settings drift after firmware update was the cause - re-apply OC tuning in 5 MHz / 5 mV increments with 10-minute stability windows.

Update AxeOS to the latest release. Bitaxe Web Flasher (https://bitaxeorg.github.io/bitaxe-web-flasher/) handles this in two minutes. Several thermal-management and TPS546-config bugs have been fixed across releases - an out-of-date AxeOS can produce false-cliff symptoms that vanish on update.

Verify pool side is not the cause. Check your pool's dashboard for the same Bitaxe - sometimes a pool-side reset of the worker config resolves what looks like a hashrate cliff but is actually a share-acceptance issue. Try a different pool (CKpool solo, Public Pool) for 30 minutes to confirm your local Ultra is producing real hashes.

Check ambient temperature at the intake with an IR thermometer at the fan inlet. Target <= 25 C. If ambient is > 30 C, the cliff may be thermally-driven without paste failure - relocate the miner to a cooler space and re-test before tearing into hardware. Many late-spring cliffs are ambient, not paste.

Re-paste the BM1366 with a high-W/mK paste. Power off, wait 60 seconds. Remove heatsink (push-pin clips or 2-4 screws depending on revision). Photograph existing paste. Clean both surfaces with isopropyl alcohol 99% and lint-free wipe until mirror-clean. Apply Honeywell PTM7950 phase-change pad cut to slightly smaller than the die top (best long-term, no pump-out for 5+ years), OR Arctic MX-6 / Thermal Grizzly Kryonaut as a ~2 mm^2 rice-grain in the centre of the die. Reseat heatsink with even cross-pattern pressure - finger-tight + a quarter turn for screws, full click for push-pins. Run 30 minutes steady-state. Expected on a paste-driven cliff: BM1366 returns to nominal ~500 GH/s and ASIC temp drops 8-18 C.

Re-torque heatsink mount. If existing paste looks fine, the cause may be lift at the mount. Reseat heatsink with even cross-pattern pressure without changing the paste. On screws-with-springs: full compression of the spring is the right pressure. On push-pin revisions: full click on every pin. Re-test 30 minutes.

Restore conservative OC profile. If your tuning was aggressive (> 510 MHz, < 1180 mV), back off to 485 MHz / 1200 mV stock or 465 MHz / 1190 mV underclock + undervolt. BM1366 silicon-lottery ceiling drifts down over time - what was stable at month 6 may not be stable at month 18. The dollar-per-watt-per-month math wins on conservative tuning, especially with the heat as partial space-heater value in a Canadian basement.

Increment back from stock if you want to push. With confirmed stock-stable, raise frequency in 5 MHz steps with a 10-minute stability window between each step. Stop one step before HW% climbs above 1.5% or temperature climbs above 70 C. That is the chip's current silicon ceiling - different from when you bought it, and different again in another 6 months. Track it.

Verify Vin at the barrel jack or USB-PD trigger. Multimeter on DC, probe at the input connector while miner is hashing. Expected: 4.9-5.1 V. Below 4.7 V = bad cable, bad jack, bad PSU, or bad USB-PD trigger. Swap one variable at a time. Degraded Vin cascades through TPS546 and ages the chip faster - cheap to fix, expensive to ignore.

Clean the heatsink fins. Compressed air from the exhaust side, push dust out (never in). A year of dust buildup looks identical to TIM failure in AxeOS readouts. Three seconds of compressed air can save you a re-paste.

Probe Vcore at the BM1366 power pads under load. Multimeter DC mode, fine probes. Probe Vcore directly at the chip pads or output cap of the TPS546. Expected at stock: 1.18-1.22 V sustained at full chip load. AxeOS report and your multimeter should agree within +/- 20 mV. If multimeter reads 30+ mV below AxeOS report under load: TPS546 regulation drift. Continue to Step 13.

Replace TPS546B24A regulator. Fine-pitch QFN soldering - preheat plus hot-air rework station, microscope helpful. Source replacement from Mouser, Digikey, or D-Central spares (TPS546B24A is a stocked Bitaxe service part). Remove old IC with hot air at 300-320 C, clean pads with flux + braid, place new IC with flux + light hot air to reflow. Verify Vcore after reassembly. Expected: regulator restores nominal, hashrate returns to ~500 GH/s.

Probe input-side capacitors and feedback resistors around TPS546. Bulging, cracked, or burnt components in the regulator path indicate wider damage - replace as needed. Output cap ESR rise alone is enough to cause regulation drift; replace output cap with a low-ESR equivalent before condemning the IC.

Replace BM1366 chip. Last-resort hardware fix and real fine-pitch BGA rework - preheat to 150 C bottom side, top-side hot air at 310-330 C for ~30 seconds, lift chip, clean pads, reball or use a fresh chip with factory balls, place with flux, reflow. Source BM1366 chips from D-Central spares (we stock them) or aftermarket recovered-grade. New-old-stock preferable. After reflow, replace TIM, reseat heatsink, run a 1-hour burn-in. Expected: nominal ~500 GH/s returns. Bench-grade work - if you do not have a preheat station and microscope, ship to D-Central instead.

Reflow existing BM1366 (long-shot recovery attempt). If chip replacement is not an option immediately and the cliff appears thermally-driven by failed BGA solder joints (cracked joints from thermal cycling can mimic chip aging), try a no-removal reflow: flux around the chip perimeter, preheat bottom to 150 C, hot-air top side at 310 C for 30 seconds, let cool naturally. Re-paste, reseat, test. Success rate on BM1366 cliff symptoms: maybe 1 in 5 - free if you already have the equipment, and the chip was already cooked from your perspective.

Inspect under microscope for visible damage. Burnt vias, cracked traces, lifted BGA pads, scorching near the chip, capacitor bulging. Any of these = stop, document with photographs, ship to D-Central. Working blind beyond visible damage destroys boards.

Stop DIY when chip replacement is needed and you do not have BGA rework capability; TPS546 replacement failed and root cause unclear; Diagnostic Step 7 confirmed failure is on this board AND Tier 2-3 fixes have not recovered it; visible heat damage anywhere on the PCB; a second cliff returns within 30 days of a successful Tier 2 fix. Book a D-Central ASIC Repair slot at d-central.tech/services/asic-repair/.

D-Central bench process. Microscope inspection of die surface and BGA pad condition; full thermal mapping with thermocouple + IR camera; TPS546 rail scope-trace under load step transients; chip-level failure analysis; BM1366 replacement from in-house stock with factory-fresh balling or pre-balled new-old-stock; PTM7950 phase-change TIM applied; post-repair 24-hour burn-in at full nominal with telemetry log. Turnaround 5-10 business days. Canada-wide; US / international welcomed.

D-Central pioneer history. D-Central was first to manufacture the Bitaxe Mesh Stand. We made the original heatsinks for both standard Bitaxe variants and the Bitaxe Hex. We have been in the Bitaxe ecosystem since the beginning, we stock every Ultra hardware revision (204, 205, 205-Bv2, 205-Cv2), and the bench has seen this exact BM1366 cliff failure mode hundreds of times. We know the difference between paste pump-out, TPS546 drift, and silicon aging on inspection - saves you the trial-and-error of working blind.

Ship safely. Power off, disconnect, wait 60 seconds. Place miner in anti-static bag, double-box with >= 5 cm of foam on every side. Include: serial number, AxeOS version, screenshot of the cliff in the dashboard, OC profile history if available, what Tier 1-3 steps you have already tried, and your contact details. Saves us bench time, saves you money. Ship to D-Central ASIC Repair, address at d-central.tech/services/asic-repair/.

Replacement Ultra option. If the chip is cooked beyond economical repair (Tier 4 chip-level repair is $120-220 CAD; refurbished or new Ultra board is in the same range), consider a replacement instead. We stock new-build Ultras and refurbished units from the bench at d-central.tech/product-category/bitaxe/. Either path keeps you mining; the math depends on whether your unit has sentimental or block-find historical value.