Goldshell – Random Crashes / Reboots Every Few Days

Power-cycle once cleanly and let it run a 24-hour observation window. Unplug the 12V brick. Wait 60 seconds. Re-plug. Re-establish pool connection. Then leave it alone for 24 hours and just watch the uptime counter in the dashboard. Some "random reboot" reports turn out to be a single one-off after a power blip, not a chronic fault. Confirm chronicity before chasing root cause - one reboot per week on a miner running 24/7 is below most operators' annoyance threshold and not always worth chasing. If reboots are >1/day, proceed.

Update firmware to the latest stable release - over ETHERNET, not WiFi. Go to System → Network and remove WiFi config. Plug Ethernet. Download the correct `.cpb` firmware for your exact model from the official Goldshell firmware page or GitHub firmware repo. System → Update → upload `.cpb`. Wait for the reboot. Verify the new version in System Info. Watch uptime for 48 hours. Most KD-BOX, CK-BOX, LT5, HS5 reboots end here. Caveat: if you are already on the latest stable and still rebooting, do not "upgrade" to a beta - downgrade to the prior stable instead.

Move the miner to a clean dedicated 240V / 120V circuit on a UPS. Random reboots from PSU sag are often caused by sharing a circuit with a compressor, electric kettle, microwave, or heat pump that pulls 800-1500 W transients. Plug a known-good 500-1000 VA UPS between the wall and the Goldshell PSU. The UPS smooths transients and outright bridges <30-second outages. CAD $120-220 once. If you cannot add a dedicated circuit immediately, the UPS alone resolves a meaningful share of these. Watch uptime for 48 hours.

Vacuum the heatsink intake and exhaust mesh. Power off, unplug. Take a small vacuum or compressed air can to the intake and exhaust grills. Don't disassemble - this is the no-tools cleanup. KD-BOX, HS-BOX, ST-BOX, Mini-DOGE accumulate dust quickly because they are often desktop-side. KA-BOX and AL-BOX II run hottest and are most thermal-sensitive. A 5-minute clean every 90 days is good preventive practice. If exhaust temp drops by >5 °C after cleaning, thermal was at least a contributor.

Reduce hashrate / tune voltage in the dashboard (if your model supports it). KD5, KD6, LT5, LT6, CK5, CK6 expose a `tuning` or `mode` selector in the web UI - try `Eco` mode or a step-down preset. This drops chip core voltage and core frequency by ~5-10%, sacrificing 5-15% hashrate for stability. If a previously chronic-rebooter holds 24-hour uptime in Eco mode, the chip cores or buck converters are operating right at the margin and Tier 2 hardware work will fix it permanently.

Multimeter the 12V barrel jack under load. Set DMM to DC volts, probes to barrel center pin (positive) and outer ring (negative). With miner hashing, expected `12.0 ± 0.2 V`, stable. Tap the PSU body lightly - if voltage wobbles, the PSU has a cold solder joint or failing cap and is on borrowed time. Note the reading. If wandering or sag below `11.6 V`, PSU is the prime suspect for Step 7.

Hot-swap the PSU with a known-good 12V brick sized correctly for your model. Ratings: KD-BOX needs 70 W (12V / 6A); KD-BOX Pro needs 250 W (12V / 21A); Mini-DOGE I/II/III+ needs 100-130 W (12V / 8-11A); KD5 needs 850-950 W (PSU is internal on most KD5/KD6/LT5/LT6/CK5/CK6/HS5 - the swap there is the internal PSU module). Use a Mean Well, Delta, or Goldshell-OEM-replacement brick - generic Aliexpress 12V supplies are not appropriate for ASIC duty. Run 24-48 hours under the swap PSU. If reboots stop: confirmed PSU.

Open the chassis and re-seat all internal cables and connectors. Power off, unplug, leave 5 minutes for caps to drain. Phillips #2, 4-8 screws. Lift the lid carefully on KD-BOX / HS-BOX / Mini-DOGE family; on KD5 / KD6 / LT5 / LT6 / CK5 / CK6 / HS5 the disassembly is more involved and the hashboards lift out separately. Disconnect and re-seat: hashboard ribbon cables, fan PWM connector, control-board power connector, internal PSU module connector. Wiggle each connector before re-seating to confirm none is loose. Reassemble. Catches vibration-loosened connectors that cause intermittent power glitches.

Inspect, clean, and re-paste the heatsink interface. Disassemble enough to lift the heatsink off the chips. Old paste shows as cracked, dried, gritty. Wipe with 99% isopropyl + lint-free cloth. Apply a pea of fresh thermal paste (Arctic MX-6, MX-4, or Kryonaut). Reassemble. This is most impactful on miners >2 years old where the original paste has dried out. Drops chip core temperatures by 5-15 °C, gives the buck regulators thermal headroom, and ends "thermal foldback into watchdog" reboot patterns.

Replace fans if RPM is below nameplate or bearings are noisy. Each Goldshell model has a specific fan part: KD-BOX uses a 4015 12V fan; Mini-DOGE uses a smaller 4010; KD5 / KD6 / LT5 / LT6 / CK5 / CK6 / HS5 / KA-BOX use 12038 12V high-static-pressure fans (typically Sunon, Delta, or Nidec equivalents). A failing fan delivers 60-80% of nameplate airflow, the chip core temp creeps up, and you are back at thermal foldback. Genuine replacements: $15-50 CAD per fan. D-Central stocks Goldshell-spec fan replacements.

Move the miner to a quality unmanaged or simple managed switch port and verify cable. Cat5e or Cat6 patch in good condition (no kinks, no crushed ends, length <15 m for a Goldshell). Plug into a known-good switch port. Avoid: PoE-injecting ports, ports with VLAN tagging, ports on managed switches that have been "secured" with port-isolation. Goldshell's network stack is fragile and we have seen miners reboot when a managed switch's MAC-table churn dropped frames. Run 24-48 hours and see if reboots stop.

Open the chassis, identify electrolytic capacitors on the input/output of the buck converters, and ESR-test each. Use an ESR meter (BSIDE ESR02 or equivalent, $40-80 CAD). Expected ESR varies by cap value but typical low-ESR aluminum-polymer or aluminum-electrolytic on a buck output is `0.02-0.10 Ω` when fresh. Caps reading >`0.30 Ω` are end-of-life. Note part numbers, voltage rating (typically 16 V or 25 V), capacitance (typically 220 µF, 470 µF, 820 µF, 1000 µF, 1500 µF). Source replacements from Digikey or Mouser - polymer-aluminum (Panasonic FP series, Nichicon PCJ) outlast standard aluminum. Replace any cap reading high ESR.

Reflow / re-solder the buck-converter switching transistors and inductors if you suspect cold solder. Hot-air rework station, flux, leaded or low-temp lead-free solder. Apply flux to suspect joints, hot-air at 280-320 °C, watch the joint reflow. This catches cold-solder failures from manufacturing or thermal-cycling fatigue. After reflow, re-test the 12V rail under load - it should be cleaner. Risk: damaging adjacent components if you wave the heat gun around indiscriminately. Skip if you do not have rework experience.

Cross-flash an alternate Goldshell firmware variant if one exists for your model. Some Goldshell models (KD-BOX, KD-BOX Pro, KD5) had multiple firmware lineages with different bug profiles. Check the GitHub firmware repo for variant releases. Cross-flash via web UI System → Update with the alternate `.cpb`. Caveat: this is exploratory and only worth attempting after firmware-update-to-stable failed. NOT recommended on KD-BOX II, KD-MAX, LT6, CK6 where firmware lineage is single-source. DCENT_OS does NOT run on Goldshell silicon - wrong hardware - so do not look for a third-party firmware port. There isn't one.

24-hour burn-in after each Tier 2 / 3 fix before declaring success. Reconnect to pool. Monitor uptime, hashrate, exhaust temperature, and event log for a full 24 hours. Random reboots are by definition stochastic - a 4-hour test isn't long enough to catch a "every 8 hours" pattern. Don't ship a "fixed" miner back to your customer or back to the rack until it has held 48 hours of clean uptime. Preferably 72.

When to stop DIY. Tier 4 territory: (a) PSU swap, firmware update, thermal cleanup, and Eco mode all failed to extend uptime past 12 hours; (b) ESR test flagged 5+ caps and you do not have hot-air rework experience; (c) the miner is >4 years old and reboots have escalated from monthly to weekly to daily despite Tier 1-2 work; (d) the chip-core voltage is wandering on the bench DMM (rail noise visible). At this point, the buck converters or chip cores are aging at the silicon level. DIY can extend life but won't restore factory stability.

What D-Central does at the bench. We pull the control board, ESR-test every electrolytic, replace polymer caps as a batch (catching weakened-but-not-dead-yet caps), reflow all buck-converter switch nodes, run a 24-hour soak test under monitored bench PSU, swap fans to known-good units, replace the heatsink TIM, and burn in for 48-72 hours before shipping back. For end-of-life models (BOX-series, Mini-DOGE I), we keep an inventory of refurbished control boards as a faster fix when component-level repair stops being cost-effective. Canadian-owned, no ship-to-China nonsense, no 6-week turnaround.

Ship safely. ESD bag the control board if shipping it standalone; otherwise foam-cradle the whole miner in a rigid box. Include a note: model, serial, observed reboot interval (e.g. "every 6-12 hours"), firmware version, mains voltage (110V / 120V / 220V / 240V), age and approximate run hours, what Tier 1-2 work you have already done. Ship to D-Central HQ - typical Goldshell repair turnaround is 5-10 business days. We send a diagnostic update before charging anything beyond the fixed bench fee.