Antminer S19 – Abnormal Fan Noise

Shut down at the breaker for 30 seconds, then power back up. A full cold start clears wedged firmware state and lets you verify whether the noise is continuous or triggered by a specific operating state. Note whether the noise is present on cold start, warm idle, or only at full load — all three are diagnostic signatures that narrow the failure mode before any tool comes out.

Shop-vac the intake filter and grills. Dust on blade leading edges is the #1 cause of 'mystery grinding' on indoor-deployed S19s. Vacuum the intake filter, grill bars, and chassis exterior. Do NOT aim compressed air directly at the fans — back-pressure can spin them past nameplate RPM and back-EMF into the control board PWM driver, creating a second problem. Vacuum, never blow compressed air straight in.

Verify ambient and intake clearance. Target intake air ≤ 35 °C, with nothing within 15 cm of the front grill. No curtains, dust piles, pet bedding, or cardboard blocking airflow. A bottlenecked intake forces the fans to work harder, accelerating bearing wear and amplifying pre-existing imbalance into audible noise. Measure intake temp with an IR thermometer pointed at the front grille, not room-middle.

Dashboard RPM sanity check. Log into the miner UI → Status → per-fan RPM. All four fans should sit within ±200 RPM of each other at steady state (~6000 RPM for standard S19 / Pro / j Pro). If one is clearly lagging, you now have the fan index to target. If all four look healthy but the noise persists, the issue is likely mechanical (shroud, chassis screw, grill) rather than fan-electrical.

Power-off hand-spin each of the four fans. Breaker off, wait 30 seconds for caps to discharge, unscrew the intake grill. Spin each fan firmly by hand. Healthy: silent, coasts 6-10 seconds. Bad bearing: gritty feel, stops in <2 seconds — grit is often felt under your fingertips on the hub. Tick-per-revolution: blade contact with grill or shroud. Visible wobble at hub: cracked hub or delaminated blade. Rank worst-to-best before touching anything else.

Inspect each blade under raking light. Remove fans one at a time (4 screws per fan). Hold each blade against a bright light at a shallow angle. Look for chips, hairline cracks, embedded debris, or missing blade fragments — damage is often <2 mm and invisible under direct overhead light. Check the hub for cracks radiating from the center shaft; a cracked hub is 'replace immediately' because those fail catastrophically at speed.

Check shroud / grill mechanical interference. With the fan out, inspect the grill for bent wires or tight spots (look for shiny witness marks where a blade has been rubbing). Check every chassis screw around the fan frame — Antminer chassis screws loosen under years of vibration. Re-torque loose screws, straighten bent grill wire, reinstall the fan. This fix costs $0 and resolves roughly 9% of 'noisy fan' tickets on our bench without replacing anything.

Multimeter the fan socket under load. Unplug the suspect fan. With the miner hashing, probe the control-board 4-wire socket: +12V to GND should read 12.0-12.5 V steady. If it reads 0 V, floats, or drifts wildly, the socket or PWM driver IC is the problem — jump to Tier 4. If the socket reads clean 12 V, the fan itself is the culprit — proceed to replace. Probe PWM and TACH pins if you have a scope; a multimeter in AC-coupled mode will show PWM activity as a drifting DC reading.

Match the replacement fan part number. S19 / S19 Pro / S19j Pro stock: `CF17512MS-B` (Beri) or `QYF17525-B` (alternate supplier) — 175×175×50 mm, 12 V / 6 A, 4-wire PWM. S19 XP / S19k Pro use a revised `CF17512MS-B` with a higher RPM ceiling (~7000). NOT interchangeable with S21 fans despite identical 175×175×50 mm form factor — S21 uses a different 4-wire connector keying. Verify the exact PN on your build's fan hub sticker before ordering replacements.

Install the replacement fan, verify RPM, observe for 24 hours. Four mounting screws, plug in the 4-wire connector, 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 this miner's pre-failure baseline. If the noise returns within weeks or a second fan starts degrading, treat as a cohort-aging issue and plan a full 4-fan replacement (step 13).

Clean and re-paste thermal interfaces if Tj was elevated during the failure window. If the bad fan had been degraded for weeks before replacement, the hashboard on that side has been thermally stressed. Open the chassis, remove the heatsinks, inspect for dried thermal paste. Replace with Arctic MX-6 or Thermal Grizzly Kryonaut in a uniform thin layer. Optional but recommended if per-chain Tj had drifted more than 5 °C above baseline during the fan's failure window.

Get per-fan RPM trending and threshold alerts. Watching RPM over days is what catches a degrading bearing before the hard alarm fires; stock shows the instantaneous value only. Braiins OS+, LuxOS and Vnish publish install packages — check your exact model on each vendor's own list. DCENT_OS is D-Central's GPL-3.0 open-source Antminer firmware (per-chip HW%, autotuning, Stratum V2 on the roadmap; of those three only Braiins OS+ ships native SV2 today, and Braiins authored and open-sourced the Stratum V2 specification) and it publishes downloadable signed experimental artifacts for exactly two lanes — the Antminer S9 and the S19j Pro on Zynq/Xilinx control boards. Of the models on this page that is the S19j Pro only; the S19 and S19 Pro are named in the project with no public artifact; the S19j, S19 XP, S19 XP Hydro, S19k Pro and the S19 Space Heater Edition have no public DCENT_OS artifact and no install route. No model, S9 included, has a published witnessed install route yet — treat a DCENT_OS flash as a bench exercise on a spare unit, not a step on a producing miner. Status: https://d-central.tech/dcent-os/beta-status/. Whichever firmware you run, set the per-fan RPM alert near the fan's own rated floor rather than the stock default, so degradation shows up before the alarm does.

Coordinated 4-fan replacement for cohort-aged miners. If your S19 has accumulated 30,000+ hours of run time (~41 months continuous) and one fan just failed on bearing wear, the other three are in the same wear window. Replace all four with a matched set. Labor is nearly identical for one vs four; the second failure is already queued up. Budget CAD $100-$240 for a matched set of four `CF17512MS-B` replacements from D-Central or equivalent vendor.

Bench-test a suspect fan on a known-good 12 V source. Pull the fan from the miner, wire it to a bench PSU set to 12.0 V, let it run 30 minutes at full speed. Listen, then IR-thermometer-check the bearing housing for localized heat. If the fan runs clean and quiet on the bench but is noisy in the chassis, the problem is mechanical — shroud, grill, chassis resonance — not the fan itself. This test separates a $35 fan-replacement ticket from a 15-minute mechanical fix.

Firmware regression test for PWM curve bugs. Rare but documented: specific stock Bitmain firmware builds have shipped overly aggressive PWM curves that ramp fans to 7000+ RPM under light load, accelerating bearing wear. Check your running firmware version against Bitmain's current release notes. If you're on a build older than 2023-Q3, rolling forward to the current official release is worth 20 minutes. On a very recent build with a new noise complaint, rolling one version back may be the fastest answer.

Cable and connector path inspection. Re-seat the 4-wire fan cable at both ends — fan-side and control-board side. Inspect for bent pins, oxidation on pin faces, blackening around the socket from localized heating. Measure cable continuity end-to-end with a multimeter. Cheap, fast, occasionally the whole answer — especially on miners that have been transported recently or have had service work in the last few months.

Stop DIY and book D-Central Repair when: three known-good fans still trigger ERROR_FAN_LOST on the same socket; visible burn marks, discolouration, or melted plastic around the fan socket on the control board; multiple fans fail simultaneously with no common cause (12 V rail issue upstream); or post-fan-swap temperature and HW% don't return to baseline within 48 hours. You're now in PWM driver IC / control board / PSU rail territory. Stop before you make it worse.

Ship hashboards or full miner safely to D-Central. Anti-static bags on hashboards; full miner in original foam packaging or 5 cm foam on every side of a double-box. Include a note with observed symptoms, fan RPM history, firmware version, and contact info. The more context we have going in, the less diagnostic time we bill. Our bench process: scope the control board with a known-good test fan and bench PSU, isolate PWM driver IC failures, reflow cold solder joints, replace burnt sockets. 24-hour nameplate burn-in before we ship back.