Skip to content
ASICademy

Send feedback

What kind?
Maintenance

Preventive Maintenance & RMA

Dust is the silent killer. Cleaning schedules, the RMA workflow, deployment burn-in, spares, and the site map.

Why dust matters (the core 'aha')

Dust insulates heatsinks and clogs fans -> airflow drops -> chips run hotter -> auto-tune backs OFF frequency to stay under throttle -> hashrate falls with NO error alert. A dusty S19 can silently lose 15-25% of rated hashrate. That's 'thermal creep' - the reason cleaning is preventive, not reactive.

PM checklist

  • Monthly: inspect grilles and pre-filters and compare fan telemetry and hashrate to baseline. Cleaning or replacement requires the approved isolation/cleaning procedure below.
  • Use a site/OEM cleaning interval based on contamination; internal cleaning requires authorized isolation and the approved cleaning, air-pressure, guarding, PPE and ESD controls.
  • Every 12-18 months (24/7 units): re-paste chips (dried TIM can cost 10-15 °C on older units).
  • Staggered scheduled reboots clear stuck cgminer/pool/memory states without dropping a big hashrate block.
Use the approved isolated-equipment cleaning procedure and ESD controls. If compressed air is permitted, use regulated cleaning pressure below 30 psi under OSHA reference conditions, effective chip guarding and PPE. Secure fan blades as the OEM specifies; switching power off alone does not establish safe isolation.

The RMA workflow

  1. Confirm it's a HARDWARE failure (rule out cable/power/config/airflow); capture the exact error string.
  2. Check warranty by serial (the white label near the LAN port). Watch typo traps: O/0, I/1, B/8.
  3. Document: serial, model, purchase proof, fault + error string, photos.
  4. Tag & quarantine the unit (fault + serial + date); pull it from the production row.
  5. Open the vendor ticket and WAIT for a written RMA number (unauthorized returns are refused).
  6. Ship (you pay inbound); keep tracking, update the ticket.
  7. Track credits in a fleet RMA ledger: RMA#, ship date, resolution; reconcile on return.

Deploy, burn-in, decommission

  • Provision each unit: rack+cable -> assign & RECORD the IP -> flash the fleet-standard firmware -> set pool + standard worker name -> bind a location tag (row/rack/position) to serial+IP -> enable monitoring.
  • Burn-in: a 24-hour soak under real conditions catches infant-mortality faults. Capture a baseline (hashrate, per-board, chip count, temps, fan RPM, HW%). Promote only if clean and within ~5% of rating.
  • Decommission: power down, de-rack, WIPE config/pool creds, remove the location tag, grade (resale/spares/e-waste), free the IP/position.

Spares & the site map

  • Stock swap items only: fans, PSUs, control boards, ribbon/data cables, thermal paste, recovery SD cards. (~10 miners: 2-3 fans/model, 1 PSU per 5; 50+: add spare hashboards.)
  • Give every slot a stable coordinate (ROW-B/RACK-04/POS-12) bound to serial + IP + worker name in ONE record.
  • Then an alert names a worker -> you walk straight to it; swaps/RMAs update one record; you can spot positional hot-rows.