Antminer S19 – Fan Grinding Noise

Breaker-cycle and re-listen on cold start. Power down at the breaker for 30 seconds, then back up. If grind is loudest on cold start and quiets as chassis warms, you have the classic cold-grease-viscosity bearing signature. If grind emerges only after 20-30 minutes of hashing, thermal expansion is worsening a marginal bearing. Either way the fan is failing; log the onset pattern — it helps confirm bearing failure vs intermittent electrical and narrows your action window.

Shop-vac intake filter and grill without spinning the fans. Dust cake on blade leading edges is the #1 accelerator of bearing wear on indoor S19s — higher RPM under same thermal load eats grease life faster. Hold each fan hub still with a soft pen while vacuuming. Never aim compressed air directly at the fans; reverse-spin back-EMF can damage the PWM driver on the control board. Vacuum weekly in pet / dusty environments, monthly minimum elsewhere.

Verify ambient intake temperature. Point an IR thermometer at the front grill, not room-middle air. Target less than or equal to 35 degC for standard S19 / S19 Pro / S19j Pro; less than or equal to 40 degC for S19 XP and S19k Pro. Every 5 degC above 35 degC ambient roughly halves remaining bearing life on any fan already past the midpoint of its L10 window. If you cannot fix the ambient, you cannot save the bearing — you can only delay the inevitable swap.

Dashboard RPM snapshot, identify the outlier. Open the miner UI (`http://<miner-ip>`), navigate to Status, read per-fan RPM values at full hash. Record all four. A fan reading 300+ RPM below its three peers at steady state is the grinder, even if noise is hard to localize by ear. This 60-second check narrows the work window before any disassembly and gives you the fan index to target.

Power-off hand-spin test on all four fans. Breaker off, wait 30 seconds for caps to discharge. Unscrew the intake grill (4 screws). Spin each fan firmly with your fingertips on the hub. Healthy fan: silent, coasts 6-10 seconds. Bad bearing: gritty feel through your fingers, stops in under 2 seconds, occasionally audible crunch as rotor decelerates. Rank the four fans worst-to-best on grit and coast time before you touch the fan mounting bolts.

Inspect each fan hub for seal failure and oil migration. Remove each fan one at a time (4 screws each). Hold the hub under a bright raking light. A healthy sealed bearing is clean and dry. A failed seal shows oil mist, dark spotting, or visible grease migration on the hub or the chassis skin directly around the fan mount. Oil out of a sealed bearing equals terminal; no amount of cleaning reverses this because the grease the bearing needs is physically gone.

Measure fan socket voltage under load. With the suspect fan unplugged and the miner hashing, probe the 4-wire socket on the control board: `+12V` pin to `GND` should read 12.0-12.5 V steady. If the reading floats, drifts, or reads zero, the PWM driver IC or socket itself is the problem — escalate to Tier 4. If the socket reads clean 12 V, the fan is the problem and you proceed to replacement. This step separates a CAD $35 fan swap from a CAD $150+ control-board repair before you order parts.

Match the replacement fan part number exactly. S19 / S19 Pro / S19j Pro stock: `CF17512MS-B` (Beri) or `QYF17525-B` (alternate supplier) — 175x175x50 mm, 12 V nominal / less than or equal to 6 A rated, 4-wire PWM with TACH feedback. S19 XP / S19k Pro use a higher-RPM-ceiling variant of `CF17512MS-B`. NOT interchangeable with S21 fans despite identical form factor — S21 uses different 4-wire connector keying. Read the PN directly off the hub sticker of your failed fan before ordering.

Swap the grinding fan, verify RPM, run 24-hour burn-in. Four mounting screws, 4-wire connector. Confirm airflow direction — the arrow on the fan frame shows rotation; intake fans pull, exhaust fans push. Close the grill. Power up. Dashboard RPM should climb to ~6000 and hold within +/-200 RPM of the three peer fans. Monitor `HW%` and per-chain temperature for 24 hours; both should return to pre-failure baseline. If new fan RPM is still 300+ below peers, return to step 7 — you have a socket or PWM problem, not a fan problem.

Inspect and re-torque all chassis fasteners while you are in there. A loose chassis or fan-frame screw that lets the fan vibrate accelerates bearing wear on that fan. Re-torque every visible screw on the fan grill, the chassis skin, and the fan frames. Takes 90 seconds, occasionally saves the next fan on this miner. While you are at it, check cable routing — any fan cable rubbing against a sharp chassis edge can chafe through over thousands of thermal cycles.

Clean and re-paste hashboard thermal interfaces if Tj was elevated during the fan failure window. If the grinding fan ran degraded for weeks before replacement, the hashboard behind it has been thermally stressed — dried thermal paste, accelerated capacitor aging. Pull the hashboard, remove the heatsinks, clean old paste with 99% IPA and lint-free wipes, apply Arctic MX-6 or Thermal Grizzly Kryonaut in a uniform thin layer. Optional but strongly recommended if per-chain Tj had drifted 5+ degC above this miner's historical baseline during the grind window.

Flash DCENT_OS for per-fan bearing-trend telemetry. DCENT_OS is D-Central's own open-source Antminer firmware — built by Mining Hackers, maintained publicly on GitHub, no licensing fees. Exposes per-chip `HW%`, autotuning, stratum v2, and per-fan RPM trending with configurable alert thresholds. After flashing, set the per-fan RPM alert at ~4500 RPM (not Bitmain's ~1500 RPM floor) to catch the next degrading bearing 30-60 days before the hard alarm would fire. Alternatives: Braiins OS+, LuxOS, Vnish — all solid; DCENT_OS is the open-source Mining Hacker pick.

Coordinated 4-fan replacement for cohort-aged miners. If the failed fan is on an S19 past 30,000 operational hours (~41 months continuous), the other three fans are in the same narrow wear window — typically 30-180 days behind the first failure. Replace all four with a matched set in a single service visit. Labor is roughly 10 minutes more for four fans vs one; parts cost CAD $100-$240 for a matched set of four `CF17512MS-B`. Buys back ~30,000 hours of uninterrupted hashing on a single visit.

Bench-verify the dead fan on a known-good 12 V source before assuming bearing. Pull the grinding fan from the miner, wire it to a bench PSU set to 12.0 V (TACH and PWM pins floating is fine for a spin test), run it 30 minutes at full speed. Listen for the grind with the fan off-chassis. Check bearing-housing temperature with an IR thermometer — a dying bearing runs 10-20 degC hotter than the housing plastic around it. If the fan is silent and cool on the bench but grinds in the chassis, you have a mechanical resonance or mounting issue, not a bearing — check fan frame mounting, chassis screws, and shroud alignment.

Firmware regression test for PWM-curve bugs. Some stock Bitmain S19 firmware builds before 2023-Q3 are community-reported to ship overly aggressive PWM curves that ramp fans to 7000+ RPM under light load — not documented in Bitmain release notes. Check your running firmware version. If older than 2023-Q3, roll forward to current official. On a very recent build with a new noise complaint, roll back one version and observe for 48 hours. Always verify specific build behaviour against a second source before trusting a single community report.

Retire the dead fan, log the hours. Record the fan PN, hub-sticker serial and date code, and operational hours at point of failure (miner uptime). Over time this builds your own bearing-life dataset for your deployment environment — a dozen data points across a couple years and you know your own L10, not Bitmain's generic L10. Your data, your horizon, your planning. This is what monitoring your own hardware on your own firmware earns you over time.

Stop DIY and book D-Central Repair when: three known-good replacement fans all fire ERROR_FAN_LOST on the same socket (PWM driver or socket problem); visible burn marks, discoloration, or melted plastic around the fan socket (stop immediately); multiple fans fail simultaneously with no common cause (upstream 12 V rail); post-fan-swap temperature and HW% don't return to baseline within 48 hours; or capacitor bulging or burnt-electronics smell near the control board. Power off at the breaker, pack for shipping. Pushing past these signatures typically turns a CAD $150 repair into a CAD $400+ control-board swap.

Ship safely to D-Central, include diagnostic context. Pack in original foam insert or double-box with 5+ cm foam on every side. Hashboards-only shipment: anti-static bags. Include a written note listing: grind onset date, RPM history for each of the four fans, firmware version, miner operational hours, ambient deployment conditions, any error codes from `kern.log`. Our bench process: scope the control board with a known-good test fan and bench PSU, isolate PWM-driver failures, reflow cold solder joints, replace burnt sockets, then 24-hour burn-in at nameplate on a loaded test rig before shipping back. The more context we have going in, the less diagnostic time we bill.