Antminer – Loose Heatsink Rattle

Power the miner off at the breaker and wait 2 minutes. Caps on Antminer PSUs hold charge, and a cold miner makes a loose-heatsink rattle distinguishable from fan noise. Set up a clean, well-lit workspace with an anti-static mat if available. Do not skip the 2-minute wait — it protects you and it separates mechanical rattle signatures from electrical hum.

Perform the tilt test before opening anything. Lift the chassis 5-10 cm off the shelf, tilt side-to-side, then front-to-back. Listen for a clear metallic slide, clunk, or roll. If you hear nothing, back out — this may not be the right error; check our fan-bearing, fan-blade, or PSU-buzzing pages first instead of opening the lid for no reason.

Remove the top-cover screws (Phillips #2 on most models, Torx T10 on some S19/S21 SKUs) and run a low-suction shop-vac over the visible interior. Keep the nozzle 5 cm from PCB surfaces. You are looking for foreign debris (zip-tie tails, packing fragments) and for any detached heatsink that has rolled loose inside. Photograph anything you find before removing it.

Under a bright flashlight at oblique angle, examine every hashboard. A seated heatsink sits square and flush with its ASIC; a lifted one shows a visible air gap, tilt, or silver/grey adhesive residue around its base. Document every anomaly with a phone photo — these become evidence if the miner is under warranty or if boards ship to D-Central.

Lightly press each suspect heatsink with a finger (miner still powered off). A bonded heatsink is rigid. A lifted one rocks 1-3 mm. Mark every loose position with painter's tape. Note any burn marks, scorched silkscreen, or visible damage on the PCB adjacent to the loose heatsink — that tells you whether the chip underneath already cooked under previous power-on cycles.

Remove the affected hashboard from the chassis. Phillips #2 for hashboard-to-chassis screws on most models; S21 mixes Phillips and Torx T8. Disconnect the data ribbon and PSU-to-board connector first, then lift the board straight up onto an anti-static surface. Photograph both sides of the board before further handling — establishes a before-state for the repair log.

Clean the ASIC top surface and the heatsink base with 99% isopropyl alcohol and lint-free wipes. Scrape any crumbled adhesive residue with a wooden or plastic pick — never metal on a bare die. Wipe until both surfaces are mirror-clean. This prep step is the single biggest determinant of whether the re-bond holds for another 2 years or sheds in 3 months under thermal cycling.

Inspect the ASIC die under a 10× loupe or phone macro. Look for cracks, chipped die corners, or thermal discolouration around the perimeter. A die that has run hot-to-failure shows subtle browning on the edges. Any physical die damage is Tier 4 — thermal re-bond cannot save a cooked chip; ship the board to D-Central instead of wasting adhesive.

Check chain continuity with a multimeter while the miner is powered off. Probe VDD to GND across the chain in question. A healthy chain reads a few hundred ohms to a few kΩ depending on model; a short (0 Ω or near zero) means a chip has failed internally and the loose heatsink is a downstream symptom, not the cause. Short = stop, Tier 4.

Re-seat the suspect heatsink dry (no adhesive) to verify physical fit. It should sit flat on the ASIC with zero rock. If it does not seat flat, the die or PCB has warped and a field re-bond will fail within weeks — ship the board. If it seats flat with no rock, proceed to the Tier 3 adhesive re-bond.

Apply fresh thermal adhesive — Arctic Alumina Thermal Adhesive (two-part epoxy) or Akasa 449 are the community-standard choices for Antminer field re-bonds. Thermal paste (MX-6, Kryonaut) is NOT a substitute; it has no structural bond. Apply a thin, even layer across the chip top — coverage, not thickness. Excess epoxy squeezes out under pressure and insulates rather than conducts, hurting the repair you are trying to make.

Seat the heatsink and apply even flat-load pressure during cure. Respect the manufacturer's full cure window — Arctic Alumina: 5 min handling, 5 hr full cure; Akasa 449 similar. Use a flat weight (a book or small sandbag) to load evenly. Do not clamp asymmetrically or you will cure a tilted bond. A rushed 30-minute cure sheds under the first week of thermal cycling.

S17-class specific: evaluate a full radiator replacement instead of per-chip re-bond. S17 / S17 Pro / T17 used soldered heatsinks on many production runs — harder to field-repair, more prone to multi-point failure. Third-party one-piece aluminium radiators (screw-fixed, not soldered) from vendors such as thanosmining eliminate this failure mode for ~$40-160 CAD and 45 minutes of swap time per board.

Flash DCENT_OS (D-Central's open-source Antminer firmware — recommended for this verification) and run the miner at stock profile for 30 minutes. Record per-chip temperature data and compare the previously-hot cluster to its neighbours. Target: less than 3 °C delta chip-to-chip within a board. Stock Bitmain firmware cannot show this. Alternatives: Braiins OS+, LuxOS, Vnish — all four expose per-chip visibility.

While the board is open and you have thermal adhesive on hand, refresh thermal paste on all neighbouring chips. If one heatsink let go because the adhesive aged past its limit, its neighbours on the same board are on the same age curve. Refreshing during the same service visit extends hashboard life and prevents a repeat teardown in 60 days when the next bond fatigues.

Stop DIY and ship to D-Central if you see any of: visible die damage, short continuity on the affected chain, a heatsink that refuses to seat flat, bulging capacitors / cracked MLCCs near the failed bond, or a re-bonded heatsink that fails again within 30 days. These indicate failures beyond field-repair scope — test-fixture, per-chip isolation, and programmable-load verification are required.

D-Central bench process: programmable load test fixture, per-chip isolation with Bitmain test binaries, chip replacement with graded BM1362 / 1366 / 1368 / 1398 stock if needed, controlled reflow and factory-grade thermal re-bond, and a full 24-hour nameplate burn-in before the board ships back. Turnaround 5-10 business days. Canada / US / international welcomed.

Ship safely. Pack each hashboard in an individual anti-static bag. Double-box with at least 5 cm of foam on every side. Immobilize internal components so nothing rattles during freight — a board that moves in its box arrives worse than it left. Include a note with observed symptoms, model and revision, firmware version, and shipping history. Honest provenance saves diagnostic time, which saves you repair dollars.