Antminer S19 – Temperature Too High (Temp1/Temp2 > 95°C)

Controlled cooldown and baseline. Power off at the PSU switch (not just the miner-side button) and leave the S19 untouched in open air for 15 minutes. Boot, run 10 minutes of steady-state hashing, and record per-chain PCB, Temp1, and Temp2 at each minute. If temps climb back to trip point within this window you have a reproducible cooling fault — not a transient. Write the numbers down and photograph the dashboard; they are the baseline for every subsequent fix.

Blow out the intake, fans, and fin stacks outdoors. Use canned compressed gas or a shop compressor capped at ≤90 psi. Spray through the intake grille, across all four fan blades, and down into the heatsink fins from both ends of the chassis. Hold each fan blade stationary with a plastic pick — free-spinning axial fans off compressed air damages the bearings. Flip the chassis and blow backward through the exhaust to clear the compacted dust at the fin base. Expect a visible dust cloud leaving the miner.

Verify ambient at the intake grille itself. Place a trusted digital thermometer directly in front of the S19's intake face, 2–5 cm off the surface, let it equilibrate 5 minutes. Reading must be ≤30 °C for continuous operation, ideally ≤25 °C. If the thermostat says 21 °C but the intake reads 34 °C, hot-air recirculation is your fault — the miner's own exhaust or a stack-mate's exhaust is being pulled back in. Re-position, add a duct, or separate stacked units with a baffle.

Clear the exhaust path. S19 exhausts at roughly 11,000 BTU/h. Blowing into a wall, corner, curtain, or another miner's intake creates back-pressure that drops airflow through the fins. Maintain ≥30 cm clearance on exhaust and make sure nothing soft (curtains, cardboard, foam) can be sucked against the intake. Re-observe per-chain temps for 15 minutes after correcting geometry — a 5–10 °C drop means airflow geometry was the fault.

Factory reset to stock autotune. Per Bitmain, the S19 reset button is active only 2–10 minutes after boot. Power on, wait 2 minutes, press and hold Reset for 5 seconds. The miner auto-restarts after ~4 minutes on factory settings. Reconfigure pool/worker afterward. This strips off any aggressive autotune or custom frequency profile pushing silicon past its personal ceiling. If temps stabilize on stock vs tuned, your tuning was the cause — rebuild it slower in Tier 2.

Replace a failed chassis fan. If diagnostics flagged a stalled fan or dead FG (tachometer) wire, swap it. The S19 uses Bitmain-spec 12 V, 4-pin PWM+FG, high-static-pressure axial fans. Do NOT substitute a quiet PC fan — the S19 needs static pressure, not low-noise airflow. Power off at the wall, remove top-cover screws, unplug the fan from the control-board header, unscrew from its frame, install the replacement in the same airflow orientation (arrow toward exhaust), reconnect, boot, verify RPM on the dashboard.

Reseat hashboard data and power connectors. Oxidation on ribbons or power blades creates marginal electrical contact, forcing voltage regulators to push harder and dumping extra heat into the board. Power off, unplug at the wall, wait 5 minutes for caps to drain. Disconnect and reconnect every ribbon and every power blade on all three hashboards. If any contact looks dull or discoloured, wipe with 99% isopropyl on a lint-free swab, let dry fully, reassemble. Zero-cost fix that resolves a non-trivial slice of thermal complaints on recently shipped or re-racked units.

Rebuild tuning slowly from stock. If Diagnostic Step 8 pointed at profile regression, don't just restore the old profile. Baseline stock (nameplate TH and wall watts), then add +50 MHz or one frequency step at a time with a 10-minute stability window between steps. Watch per-chain temps and HW%. Stop at the step before chip temps cross 90 °C sustained or HW% crosses 2%. That's this S19's personal silicon-lottery ceiling — differs per miner, per board, per season. On DCENT_OS set a per-chip temp alarm at 92 °C to flag drift before the trip.

Swap suspect hashboard between slots. If Diagnostic Step 4 isolated one bad board, swap it into a known-good slot and a known-good board into the suspect slot. Run 20 minutes each. If the fault follows the board, the board is the problem (continue to Tier 3 — repaste, reflow). If the fault stays in the slot, the slot is the problem (fan zone, control-board header, local recirculation) — revisit Steps 6 and 7 targeted at that slot.

PSU swap test. If diagnostics flagged rail sag, install a known-good APW9 or APW12 matched to your S19 revision (check Bitmain's compatibility matrix at support.bitmain.com/hc/en-us/articles/4809007246617 before mixing PSU families). Run 30 minutes loaded, re-check temps and HW%. Persistent sag on a healthy PSU means your circuit is undersized or line voltage sags — electrician territory, not repair. A dedicated 240 V / 20 A branch fixes this category completely.

Repaste one or all three S19 hashboards. Highest-value advanced fix on any S19 at 18+ months of continuous service. Power off at the wall, wait 30 minutes. Remove hashboard from its slot, tracking connector orientation. Remove heatsink retention hardware; warm stubborn paste with a hair dryer on low for 30 seconds — never pry with metal. Clean every BM1398 / BM1362 / BM1368 top and mating heatsink face with 99% isopropyl and lint-free wipes until shiny. Apply a rice-grain dot of Arctic MX-6 or Thermal Grizzly Kryonaut per chip. Replace crumbled VRM / PMIC thermal pads with fresh same-thickness pads (typically 1.0 mm or 1.5 mm — measure the originals). Re-seat heatsink using original hardware in a diagonal torque sequence over two passes. Reinstall, boot, observe. Expect 5–15 °C drop on the repasted chain once paste cures 15–30 minutes under load.

Reflow a single hot chip. If diagnostics isolated one BM1398 / BM1362 / BM1368 position running 15 °C+ hotter than neighbours after a clean repaste, the BGA joints have fatigued. Remove the heatsink, apply flux to the chip perimeter, preheat bottom side of the board to ~150 °C, top-side hot air at 310–330 °C for ~30 seconds. Cool naturally (no fan, no solvent on hot PCB). Repaste, reassemble, re-run. BM13xx BGA packages tolerate a reflow cycle well — lowest-risk chip-level intervention on an S19-class board. If the same position fails again within 30 days, or the same position appears on a second board, escalate to Tier 4.

Replace a failed NTC temperature sensor. If Diagnostic Step 7 showed sensor fault (255 °C at cold-boot, stuck 0 or -40 °C at cold-boot), this is through-the-board SMD rework. Hot-air rework station, flux, replacement NTC bead (S19 family uses 0402 or 0603 NTCs, typical 10 kΩ @ 25 °C, B = 3950 — measure your specific footprint and confirm exact part number against the hashboard silk-screen or Bitmain's hash-board-code-editor output before ordering). De-solder the failed bead with narrow hot air, clean pads, drop the replacement with tweezers, reflow. Advanced work — stop and ship to Tier 4 if you're not already comfortable with SMD rework on a dense mining board.

Re-flash the PIC on a hashboard reporting `fail to read pic temp`. If kern.log shows `fail to read pic temp for chain X` before the over-temp trip, the PIC microcontroller has lost its program or the I2C line to the control board has degraded. Bitmain's hash-board-code-editor tool can re-flash the PIC; not widely distributed outside the repair community but documented on Bitmain's community forum (support.bitmain.com/hc/en-us/community/posts/900002377066). If the re-flash brings sensor reads back to sanity, your thermal fault was actually a PIC-comm fault. Common on BHB42-family hashboards aged 2+ years.

Cross-flash DCENT_OS for per-chip visibility and tuning control. For any S19 you want to keep alive long-term, cross-flash DCENT_OS (https://d-central.tech/dcent-os/) — D-Central's own open-source Antminer firmware, with GitHub source at https://github.com/DCentralTech/DCENT_OS. DCENT_OS exposes per-chip HW%, per-chip temperature, autotuning with hard temperature caps you set, per-domain voltage control, and Stratum V2 — all the forensics that make ERROR_TEMP_TOO_HIGH diagnostics sharp. Built by Mining Hackers for Mining Hackers, open-source, no licensing lock-in. Braiins OS+, LuxOS, Vnish are viable alternatives. Back up pool/worker config before flashing anything.

Stop DIY and book D-Central repair when: (a) a single-chip hot spot did not move after a clean repaste; (b) the same chip position fails on two different boards in the same rig; (c) a reflowed chip fails again within 30 days; (d) visible capacitor bulging, MLCC cracks, scorch, or burnt-electrolytic smell; (e) dead sensor and no SMD rework capability. D-Central's bench runs the full S19-family fixture set — programmable load, per-chip isolation with Bitmain test binaries, salvaged-grade and NOS BM1398/BM1362/BM1368 for chip replacement, 24-hour nameplate burn-in. Ship hashboards only (not the full chassis). Pack anti-static, double-box with ≥5 cm foam, include symptoms + firmware version + contact. Mark the ticket `Antminer S19 thermal`. Book: https://d-central.tech/services/asic-repair/