Goldshell – Fan RPM Error / Loud Bearing Noise

Cut power at the wall and confirm the thermal state first. A grinding or `0 RPM` fan on any Goldshell unit in the 1500 - 3500 W class is a thermal emergency, not a maintenance window. The web UI reboot button does not cut power to the heatsink; the rocker on the PSU may not either depending on the unit. Pull the plug at the receptacle, wait 60 seconds for the chassis to start releasing heat, then open the lid. A KD6 or LT6 running 30 minutes with a dead fan can warp thermal pads and shorten hashboard life by months. This step is non-negotiable, regardless of how good your DIY skills are.

Read the dashboard, write down which fan is faulted. On the miner's web UI - reach it at `http://<miner-ip>/` or via the Goldshell discovery tool - look at the status / monitor page. Note exactly which fan slot reports `0 RPM` or shows a red icon, what the chip temperatures read, and whether the system log shows a recent `FAN_ERR`, thermal throttle, or shutdown. Screenshot the dashboard or copy-paste the log text. On multi-fan models like KD6, LT5 Pro, or HS5, the slot number tells you which physical fan to inspect, which dictates the part number to order, which dictates whether you're servicing the unit at home or shipping it.

Blow the fan stack out with canned air. Upright can, short controlled bursts at 6 - 8 inches. Hold each blade still with a plastic probe while blasting - letting canned air spin a coasting fan backward through its motor windings induces back-EMF that finishes off a tired bearing in seconds. Pay particular attention to the hub area, where dust dams and pet hair accumulate. Goldshell BOX-series miners are particularly prone to dust intake because their cases are not gasketed. A 90-second canned-air session resolves a real fraction of `FAN_ERR` tickets on its own.

Check ambient temperature at the intake. Goldshell units publish operating ranges in the `5 - 35 °C` ambient window depending on model. If your intake is sitting at `38 °C` because it's August in Toronto and the miner is in a closet, the fan governor may be pinning the survivors at 100% duty cycle - what you're hearing is the firmware doing its job around hot air, not a bearing failure. Confirm by moving the unit to a cooler space for 10 minutes and re-checking the dashboard. If RPM normalizes, the fan is fine; the room is the problem.

Reseat the fan connector at the control board. Power off at the wall, wait 60 s, open the chassis (Phillips #2 or T10 Torx depending on model). Locate the 4-pin fan header on the Goldshell control board matching the dashboard's faulted slot. Unplug the connector, inspect the shell for bent pins, oxidation, or a cracked plastic latch. Reseat firmly until you feel the click. Before closing the lid, dab a trace of dielectric grease on the pins - chassis vibration is what backed it out the first time, and grease is what prevents it from happening again. This single step resolves a meaningful slice of `Fan = 0` tickets.

Inspect the cable end-to-end. Run a finger along the entire fan harness from the fan body to the control-board header. Look for chafe points where the cable runs past sheet-metal edges in the chassis, strain-relief damage at either terminal, heat discoloration on the jacket from a nearby heatsink, or crush damage from a previous lid reinstall over a pinched cable. Replace the harness if any of those are visible. Goldshell's standard 4-pin convention is `GND / +12 V / TACH / PWM`, but verify against the silkscreen on your specific control-board revision before assuming - they have not documented pinouts publicly.

Manually spin-test each fan with power off. Plastic probe, gentle push to start each blade. A healthy axial fan spins freely for 3 - 8 seconds and coasts to a smooth, silent stop. A failing bearing makes audible grit, binds at a rotation point, stops abruptly, or starts ticking immediately. Compare the suspect fan to its siblings on the same unit - that comparison is the single most diagnostic test you can run without a multimeter. If the fan grinds in your hand, it is grinding in service; the noise you're hearing on the dashboard is real.

Measure +12 V rail at the suspect fan header under operating power. Multimeter on DC volts, miner powered on, controller commanded to drive the fan. Probe `V+` to `GND` on the control-board header. Expect `11.8 - 12.2 V` steady. If you read below `11 V` or you read `0 V`, the rail is dead - blown SMD fuse or a supply-rail fault on the control board. That's Tier 3 / Tier 4. If you read healthy `12 V` and the fan still doesn't spin, the fan motor is dead, the harness is open, or the PWM line is stuck low - swap the fan to confirm.

Swap the suspect fan into a known-good slot (multi-fan models only). KD5, KD6, KD-MAX, LT5, LT5 Pro, LT6, CK5, CK6, and HS5 all expose at least two fan headers on the control board. Power off, move the suspect fan's harness to a known-working header, power on, re-read the dashboard. If the suspect fan now reports real RPM in the new slot, the *original* slot is the problem - a board-side fault. If the suspect fan still reads `0` in a known-good slot, the fan itself is dead. This is the single most important diagnostic isolation you can do at the bench.

Check the firmware version. Note the running firmware on the dashboard. Cross-reference against open issues on the Goldshell firmware GitHub repo - certain KD-BOX and Mini-DOGE releases regress sensor and fan reporting. If you're on a known-bad release (the `2.1.1` / `2.1.3` family on KD-BOX is the recurring one), the fan reading might be a firmware side-effect rather than a hardware fault. Never flash Goldshell firmware over WiFi - their own KB explicitly requires ethernet, and WiFi-borne flashes are the #1 brick cause documented in their support tickets.

Replace the fan with a correctly-spec'd part. Goldshell does not publish official part numbers, so you size by physical dimension and electrical spec. BOX-series typically run `60 x 60 x 20 mm` or `80 x 80 x 25 mm` 4-pin axials at `12 V / 0.3 - 0.6 A`. HS5 / CK5 / KD5 use `120 x 120 x 38 mm` server-class axials at `12 V / 1.5 - 2.5 A / ~3000 - 4500 RPM`. KD6 / CK6 / LT5 / LT6 / KD-MAX use dual `120 x 120 x 38 mm` or `120 x 120 x 25 mm` axials at `12 V / 0.6 - 1.5 A` each. Match dimensions exactly. Do NOT use a 3-pin fan in a 4-pin slot - PWM control is what keeps the firmware from screaming at full RPM 24/7.

Replace the SMD fuse on the control-board fan rail. If your voltage measurement showed the `+12 V` rail dead on one fan header while others were healthy, an SMD fuse popped when a previous fan seized. Goldshell control boards typically use 1206-package fast-blow fuses in the `2 - 5 A` range on each fan-rail tap. Identify by following the rail trace from the fan header back to the nearest SMD fuse. Reflow the replacement with hot air at `290 °C`, Kapton tape protecting adjacent plastics. Verify rail voltage returns to `12 V` under a dummy load before reconnecting a real fan. If you're not comfortable with SMD rework, this is where you stop.

Replace the Goldshell control board. If the fan rail is dead and the SMD fuse swap didn't restore it, damage is deeper than a single component - dead PWM gate driver, voltage regulator fault, or board-level short. Replacing the full control board is faster than further component-level rework. Compatible Goldshell control boards are a documented repair SKU at major bench suppliers (Zeus Mining LT5 control board, CK5/KD5 control board). Match the control-board model exactly to the miner - Goldshell has not standardized cross-model board compatibility the way Canaan has on the MM3 family.

Refresh thermal paste on the hashboard if overheat events occurred. If the dead-fan state ran long enough to trigger thermal-shutdown reboots, the thermal-interface material between the ASIC dies and the heatsink took abuse. Pull the heatsink, replace TIM with Arctic MX-6 or Thermal Grizzly Kryonaut, inspect thermal pads on PMICs and temperature sensors for crumbling. This is preventative work that pays off if you want to keep a Goldshell unit operational past the warranty period - and it doubles as a root-cause fix if the original problem was thermal pad degradation rather than a fan failure.

For BOX-series miners with a stuck red+green LED after a related firmware event: if your `FAN_ERR` event coincided with a failed firmware upgrade (both LEDs solid, no web UI), you're dealing with a firmware brick on top of a fan fault. Goldshell BOX miners require an SD-card recovery flash via the procedure documented by James Chambers. The recovery image (`burn-*.img`) is not on GitHub - you have to email `hello@goldshell.com` to get it. Do not attempt fan repairs on a bricked BOX miner; recover the firmware first, then diagnose the fan fault with a working dashboard.

Stop DIY when any of these are true: the `+12 V` rail is dead at multiple fan headers (board-level fault), visible heat damage on the control board, multiple fans failed in the same rig within 30 days (upstream cause - PSU degradation or grounding fault), a fan failure that coincides with a hashboard going dark, or you attempted SMD rework and lifted a pad. That's [D-Central ASIC Repair](https://d-central.tech/services/asic-repair/) territory - we service Goldshell hardware on the same bench that handles Antminer, Whatsminer, and Avalon.

What D-Central does at the bench for a Goldshell fan event. Diagnosis against a thermal camera and a known-good reference unit, fan-rail SMD fuse and gate-driver replacement, fan harness remake with dielectric-greased connectors, full hashboard thermal service if overheat events occurred, firmware recovery and upgrade via ethernet (never WiFi), and a 6-hour nameplate burn-in with continuous fan-RPM monitoring before the unit ships back. Goldshell does not run a North American repair channel - if you're in Canada or the US and your KD6 or LT5 Pro is grinding, you're either shipping it overseas or shipping it to D-Central.

Ship the whole miner, packed thermally aware. Double-box the unit, anti-static wrap the hashboard if it's a serviceable model, include a note with: failing fan slot, dashboard screenshot, firmware version, observed symptoms, and any bench measurements you took. That saves diagnostic hours, which saves you repair dollars. Canadian customers ship to our Quebec bench and have the unit back in 5 - 10 business days. US and international welcome - we're one of the few North American benches taking Goldshell hardware as a first-class repair platform, not a side hobby.