Avalon 1266 – Temperature Too High

Pull the power cord at the PDU, wait 60 seconds for bulk caps to drain, then restart. A full cold-boot clears any wedged MM3 firmware state around the OVER_TEMP fault flag — Canaan's MM3 build occasionally holds the flag across a soft reboot. Count to 60, do not just hit the dashboard restart. This step alone clears roughly one in five events on a 1266 that saw a transient thermal spike. If the miner boots clean and runs an hour without re-tripping, you caught a transient. Re-trip within minutes — escalate.

Vacuum the front intake grille and rear exhaust vent with a shop-vac using the soft-brush attachment. Pay particular attention to the four YD12038B2G fan blades and the heatsink fin packs visible through the grille. Dust-packed fins can raise the heatsink-to-ambient delta by 8-15 °C, which is the entire margin between running fine and OT on a warm day for a 1266. Use the soft-brush; a hard plastic nozzle risks chipping blade edges.

Check the intake-air temperature with an IR thermometer or probe thermometer held 5 cm from the front grille during hashing. Target ≤ 30 °C for a Canadian basement or garage install; ≤ 32 °C to keep useful margin; ≤ 35 °C is the Canaan absolute maximum. The 1266 has less ambient margin than the 1166 or 1246 because the A3206 is binned closer to its ceiling at nameplate. If intake reads > 30 °C, open a window or move the miner's intake to a cooler corner before doing anything more invasive.

Confirm physical clearance: at least 30 cm in front of the intake, at least 15 cm behind the exhaust. A 1266 piled into a shelf with no breathing room recirculates its own exhaust — which is exactly how a miner that ran fine last week starts throwing OVER_TEMP this week after someone stacked a box in front of it. Clearance is free; use it. Confirm airflow direction is front-to-back if the chassis was reinstalled after a service event.

If you run multiple miners in a shed, garage, or basement rack, reconfirm airflow direction. Miners aimed at each other's intakes cook each other. Correct 1266 layout is cold intake from the front on a dedicated cold-air plenum, exhaust to outside or into a heat-utilisation duct for home heating. Recirculating exhaust into intake is the fastest way to hit OT in any install, and the 1266's lower thermal margin makes it the first miner in the rack to fail when this happens.

Open the chassis, remove each of the four YD12038B2G fans in turn, and spin them by hand. A healthy dual-ball-bearing fan spins freely for 2-3 seconds after you flick the blade. A failing fan grinds, buzzes, or stops immediately. Replace any fan that fails this test with a drop-in YD12038B2G (or 12038-class equivalent) — the 1266 uses the same fan as the 1066 Pro / 1126 / 1146 / 1166 Pro / 1246 per Zeus Mining's part listing. Do NOT substitute a 120×120×25 mm PC case fan; the chassis needs the 38 mm depth for static pressure.

Measure the 12 V fan rail at the PSU-to-control-board connector under load. Expect 11.8-12.2 V sustained. Below 11.6 V means a tired PSU sagging the fan rail — fans slow, chips heat, miner trips OT. Swap to a known-good PSU from the A11/A12 product line before assuming the miner is the problem. Verify line voltage at the wall under load: 235-245 V on 240 V split-phase, 202-212 V on 208 V commercial. Low line voltage forces the PSU to draw more current and trip its own thermal sensor independently.

Re-torque the heatsink mounting clips on each of the three hashboards. Canaan's 1266 heatsink clip loosens over 12-18 months of thermal cycling; on a 1266 specifically it loosens faster than on a 1246 because the chip runs hotter. A heatsink with even 0.1 mm of lift loses the paste contact patch, and the chip underneath cooks while its neighbours are fine. Remove the heatsink, wipe the old paste with 99% IPA, apply a thin uniform layer of Arctic MX-6 or Thermal Grizzly Kryonaut, reseat with even clip pressure. Do one hashboard at a time.

Install a foam pre-filter on the intake if your install does not already have one. D-Central strongly recommends a simple foam pre-filter (3M Filtrete M8-class media) on every 1266 install — it catches dust before it hits the heatsinks, and you can vacuum or wash it in seconds instead of opening the chassis. Without a pre-filter, you're relying on the heatsink fins themselves to catch debris, which is exactly the failure mode this guide fixes after you've already ground months of wear into the fans.

Snapshot cgminer-api estats output to a text file before any further changes. Run `echo -n '{"command":"estats"}' | nc 127.0.0.1 4028` from the miner or a network-connected laptop. Save as 1266-pre-fix.txt along with the PS error integer, BOOTBY codes from kernel log, MM3 firmware build, and observed intake ambient. If the miner ends up shipping to D-Central, that pre-fix snapshot is exactly what the bench tech will want to see and saves diagnostic dollars.

Upgrade the intake duct if the install is in a confined space. A bare 1266 in a basement utility room shares its intake air with whatever else is in the room (water heater, dehumidifier, dryer venting). A short rigid duct from outdoor cold-air supply to the front grille fixes most miner-trips-at-4PM patterns instantly. Size the duct cross-section for ≥ 840 CFM design throughput, same as the miner's stock fan capability. An undersized duct chokes airflow and creates the exact problem you're trying to fix.

Remove a hashboard, strip the heatsinks, and reapply thermal paste on every A3206 ASIC on the chain that dominated your PVT_T outlier list. Use Arctic MX-6 or Thermal Grizzly Kryonaut in a thin uniform layer — the grain of rice heuristic is for CPUs, not a ~60-chip hashboard; you want a uniform film across each chip top. Replace any thermal pad under the PCH or voltage-domain ICs if the pads are crumbled, discoloured, or compressed. Budget one hour per hashboard the first time, 30 minutes once you've done a few.

If one specific chip position is the outlier and paste refresh did not resolve it, reflow that single chip. Preheat the hashboard from below at 150 °C for 3 minutes, hot-air from above at 310-330 °C for approximately 30 seconds with flux around the chip periphery, natural cool-down. The A3206 BGA tolerates a single reflow cycle reliably. A second reflow on the same chip within 90 days rarely sticks — at that point the chip is dying and needs replacement from D-Central salvaged-grade inventory.

Flash a known-good MM3 firmware build from Canaan's firmware portal at avalonminer.org/firmware-document/. The portal is login-gated and release notes are sparse, but it is the only official source for MM3 builds. Verify your hardware revision against the firmware compatibility table before flashing — the wrong MM3 build for a late-rev 1266 can brick the control board. Flash via the dashboard's firmware page over a stable wired connection; never flash over wireless. If you've hit the documented 22061301_be77c30_ef5defc sensor-polling regression, downgrade to 22033101_4ec6bb0_49ce84a or vetted earlier build.

If the fault cleared after firmware update but returned within a week, downgrade to the previous stable build and note the revision numbers. Document your working build and flash date in a text file taped to the chassis lid. The 1266 has less published community A/B firmware data than the 1246 because it had a narrower release window; expect to run your own A/B and share findings on the bitcointalk Avalon thread to help the next operator.

Reseat the control-board ribbon connectors on all three hashboards. The same IDC-style ribbons used on the 1246 MM3 show up on the 1266, and they oxidize the same way in humid environments. Pull each ribbon fully, wipe contacts with 99% IPA on a lint-free wipe, reseat until the latch clicks. Oxidized ribbons can break the temperature telemetry path specifically, causing a chain to appear cool while it's actually running hot — which then trips a different sensor and confuses diagnosis.

Stop DIY and ship to D-Central's Avalon bench when DIY is exhausted: every fan replaced, every heatsink cleaned, every chip repasted and the miner still trips OT within 24 hours; reflow returned within 30 days; PVT_T shows impossible readings; same chip position fails on multiple units; visible heat damage. D-Central bench process: test-fixture isolation per chain under controlled ambient, FLIR per-chip thermal imaging at load, graded-chip replacement from salvaged-grade A3206 inventory, full reflow with fresh paste, 24 h nameplate burn-in at 100 TH/s before shipping. Turnaround 5-10 business days, Canada-wide, US/international welcomed.

Pack for shipping: anti-static bags on each hashboard and the MM3 separately, double-box with ≥ 5 cm of foam on every side, include a printed diagnostic note with the kern.log excerpt (actual OT lines), MM3 firmware build, PSU model, measured line voltage, intake ambient, fan-tach readings before failure, and every Tier 1-3 step you've already run. The bench tech starts from your notes — the better your notes, the faster (and cheaper) the repair. Never ship hashboards loose or single-box; A3206 BGAs do not survive careless handling.