When an Antminer throws a high-temperature warning or drops to zero hashrate, it is protecting itself: the firmware holds the ASIC chips in reset and cuts board voltage to zero the moment any hashboard sensor crosses its critical threshold. The fix is almost never the miner — it is airflow, dust, thermal contact, or a lying temperature sensor. This guide covers how thermal protection works on Antminer S9/S9i/T9/L3+ and newer models, how to read the real cause from the logs, and how to bring the machine back online for good.
How Antminer thermal protection actually works
Every Bitmain hashboard carries dedicated temperature-sensor ICs (LM75/TMP1075-class parts on the I2C bus) positioned at the board’s air inlet and outlet, plus the on-die temperature diodes inside the ASIC chips themselves. Firmware polls both continuously. On an S9-generation board the chip is the BM1387; the L3+ runs the BM1485. The sensor network reports the hottest point on the board, not an average.
When a board’s reported temperature exceeds its limit — roughly 80–85°C at the hashboard on legacy models like the S9i — the controller does two things in sequence: it asserts chip reset and it commands the board’s power domain to 0 V. That is why an overheating miner reads as zero hashrate rather than a degraded one. The machine has not failed; it has parked itself. Bitmain’s environmental spec calls for an ambient intake temperature below 35°C; sustained operation above that pushes chip temperatures into protection territory even on a clean, healthy unit.
Confirm it is a thermal trip, not something else
A zero-hashrate miner has several possible causes. Confirm thermal before you start cleaning:
- Open the miner’s web UI and check the Status page for live board temperatures and fan RPM.
- Open the kernel log (Miner Status → Kernel Log) and search for high-temperature or “temp” warnings, protection events, and which board number (chain) tripped.
- If you have SSH access to the control board, you can cross-check what it exposes with
cat /sys/class/hwmon/hwmon*/temp*_input(values in milli-degrees, e.g.82000= 82°C). Treat this as a secondary check only — it reflects the control board’s aggregated sensor view rather than a specific hashboard inlet/outlet IC, so the kernel log in step 2 remains the authoritative reading.
Note which board tripped and whether all three read hot or just one. A single hot board points at that board’s cooling or thermal contact; all three reading hot at once points at ambient or airflow.
Bring it back online: the cool-down sequence
- Power down and let it rest. Cut power at the PSU and give the unit 2–3 minutes. Rebooting into the same heat just re-trips protection in a loop — if the room is still hot, the miner will keep shutting down.
- Lower the ambient temperature first. No amount of internal cleaning helps if intake air is already at 35°C+. Improve room exhaust, add negative-pressure ducting, or move the unit to cooler intake before anything else. Antminers are hot-aisle/cold-aisle machines: they must pull cool air in one end and dump hot air out the other with no recirculation.
- Clear the dust. Dust blanketing the heatsinks is the most common self-inflicted cause. Power off, remove the fan shrouds, and blow the boards out with dry compressed air — heatsink fins, fan blades, and the intake grille. Do this outside or over a bin; a season of hashcenter dust chokes airflow and insulates the chips.
- Verify both fans spin freely and at full RPM. A seized or under-speed fan starves one end of the board. Compare intake and exhaust fan RPM in the UI against each other; a large mismatch or a fan reading 0 is your culprit. Fans are a cheap, standard 12 V 6-wire part — replace, don’t nurse.
- Reboot and watch the temperatures climb. Bring it back up and monitor the Status page for the first 10–15 minutes. Healthy boards settle into a steady gradient (outlet warmer than inlet). A board that races straight back to the trip point has a hardware cause, below.
When cooling doesn’t fix it: hardware causes
If a cleaned, well-ventilated miner still trips — especially the same board every time — the fault is on the board:
- Dried or detached thermal interface. Heatsinks are clipped to the ASICs over thermal paste or pads. When that paste dries out or a clip fails, the chip loses its heat path and shuts down while its neighbors read fine. This shows as a localized hot spot under a thermal camera and is a board-level repair: reseat the heatsink with fresh thermal compound.
- Failed temperature sensor. A dead or I2C-stuck sensor IC makes the firmware think the board is overheating and trips protection falsely — the board may actually be cold. If the kernel log reports an implausible temperature (a wildly high or fixed value) on a board that is cool to the touch, suspect the sensor, not the chips.
- Chip-level thermal runaway. A degrading chip in a power domain can draw excess current and heat itself and its neighbors. This usually pairs with rising hardware-error rates and dropped chip counts in the log.
Diagnosing these means reading board temperatures per-chain, probing per-domain, and ideally scanning the board with a thermal camera to find the cold (detached heatsink) or hot (runaway) spot. Remember the miner regulates voltage per power domain, not per chip — a whole domain’s worth of chips shares one rail, so a single bad domain drags its group.
Common mistakes
- Reusing the compressed-air can indoors. Blowing dust back into the same room and re-ingesting it. Clean where the dust can leave.
- Running fans on a bench without load and calling it fixed. Thermal faults only appear under hashing heat; verify after a real reboot, not at idle.
- Ignoring intake temperature. A 40°C room will trip a perfectly good miner forever. Cooling the room is not optional.
- Swapping boards before checking the sensor. A falsely tripping sensor sends good boards to the scrap pile. Confirm the reading is real first.
Escalate to repair
If the miner still enters thermal protection after cooling, cleaning, confirmed-good fans, and fresh thermal paste, the board needs bench-level diagnosis — sensor replacement, heatsink reseating, or domain-level chip work that requires hot-air rework and a test fixture. That is a repair-shop job, not a field fix.
Related: Walk a dead or zero-hashrate board through structured diagnosis with the ASIC Fault Finder, pull the exact temperature and fan specs from the miner manuals, source replacement fans, sensors, and thermal compound from ASIC repair parts, or hand it to the bench via Start a Repair.
