Goldshell HS5 / KA-BOX – Persistent High Temperature

Move the miner. Pull it off the carpet, off the wall, out of the cabinet. Give it 30 cm clear on the intake side and 50 cm on the exhaust side. Make sure it is not breathing its own exhaust. If you have multiple miners stacked, the upper unit is always inhaling the lower unit's 50-60 C exhaust - stop doing that. Wait 20 minutes and re-check the temperature widget. A correctly relocated HS5 typically drops 3-7 C on chip temp with no other change.

Vacuum the intake grille and blow out the heatsink fins. Power off, unplug, take it outside or to a garage. Vacuum the intake grille first (a shop-vac with a brush head is ideal). Then blow compressed air backwards through the heatsink - exhaust to intake direction - holding fan blades stationary so you don't spin them up and damage the bearing. Repeat from the intake side. A first-cleaning on a 6-12 month old miner will release a visible cloud of dust. Resolves ~50% of high-temp tickets on its own.

Drop ambient by 3-5 C. Open a window, switch on a fan in the room, run a portable AC if you have one, or move the miner to a cooler space. Even a 20 cm clip-fan blowing across the intake grille adds enough air movement to drop chip temp 2-4 C. Canadian winter? Crack the window to your basement - the BOX-series loves a 15-20 C intake. The published Goldshell spec hard-caps intake at 35 C and starts throttling above 28 C ambient.

Revert any custom tuning profile. In the Goldshell web UI, go to Setting -> Mining and confirm the frequency profile is set to stock / default. If you or a previous owner pushed an overclock, revert it. Custom profiles often raise the firmware's internal temp target by 5-10 C to keep the boost stable. Reboot after reverting, let the miner re-stabilise for 30 minutes, then re-read temperatures. Stock profile + clean intake = stock temperature.

Update to the latest stable Goldshell firmware - over Ethernet only. Sign in, check Setting -> System for available updates. If a newer build is offered, plug Ethernet directly. Do not upgrade over WiFi - Goldshell's Zendesk article 16805936980633 is explicit about this and a botched WiFi upgrade is the #1 way to brick a BOX miner. Some firmware revisions have tightened the temperature target back down toward 75 C. Reboot and retest after upgrade.

Open the chassis and inspect the heatsink + thermal interface visually. Power off, unplug, ground yourself. Remove the four chassis screws (Phillips #2). Lift the top cover. Inspect the aluminium heatsink: leading edge of fins should be sharp aluminium, not grey-felted dust. Inspect the visible edge of the thermal pad / paste between the ASIC die and the heatsink base - it should be uniform, not cracked, dried, or pumped-out at the corners. Note what you see, take photos for the Mining Hackers if you escalate.

Deep-clean with brush + compressed air, paying attention to fin gaps. With the chassis open, use a soft anti-static brush along the leading edge of every heatsink fin to break up any felted dust mat. Then compressed air at 30 cm distance, multiple short bursts, holding the fan blades with your finger so the air doesn't spin them up past their rated RPM (this can damage sleeve bearings instantly). Vacuum loose debris. Reassemble and run for 1 hour. Re-check temps - well-maintained units land back in the 60-70 C band.

Verify fan health: RPM under load + bearing noise. SSH into the miner (ssh admin@<ip>, password 123456789 on stock firmware) and read fan RPM via /proc or the web UI's hardware page. Compare to the rated max printed on the fan label (HS5 stock fan is typically ~6000 RPM). If the fan is more than 15% below rated max under full load, the bearing is degraded. Listen at the exhaust side - any grinding, whining, or rhythmic ticking confirms wear. Schedule a fan swap (Step 10).

Measure the 12 V rail at the PSU output under load with a multimeter. Set DMM to DC volts, probe across the PSU output to the miner. Healthy rail measures 11.8-12.2 V under hashing load. If you see <11.7 V sustained, the PSU is sagging - a sagging 12 V rail starves the fan, RPM drops 10-20%, airflow drops, and chip temp climbs. Swap the PSU with a known-good 12 V brick rated for the miner's power draw (HS5 is ~2300-2500 W peak; HS-BOX much lower at ~150 W).

Replace the fan with an equivalent 12 V PWM unit if RPM is degraded. Source a same-size, same-airflow-rating, 12 V PWM fan. For HS5 this is typically a 120 mm server-grade fan; HS-BOX is a smaller 80-92 mm unit. Note the original connector pinout (+12V, GND, PWM, Tach) and either re-use the original connector via solder splice or crimp a new one. Reinstall, route the cable away from the heatsink, power up, confirm the new fan ramps to rated RPM under load. Chip temp drop after a fan swap on an aged unit is typically 4-8 C.

Reposition the miner relative to other heat sources and confirm intake-air temperature with a logged thermometer. Use a cheap $15 USB thermometer logger near the intake grille for 24 hours. If intake spikes above 30 C during peak room-load (other miners, sun on a wall, HVAC dead zone), you have an environmental problem the miner cannot fix. Add a duct, a baffle, or relocate. Mining Hackers running 4+ BOX-series in a single room almost always end up adding a dedicated exhaust to outside air.

Repaste the ASIC die / heatsink interface with a high-quality TIM. Unscrew the heatsink mounting hardware (typically 4x M3 screws on the BOX-series, sometimes spring-loaded). Lift the heatsink straight up - do not twist if the paste has hardened. Clean both the ASIC die top and the heatsink base with 99% IPA and a lint-free wipe until both surfaces are mirror-clean. Apply a pea-sized dot of Arctic MX-6 or Kryonaut. Re-seat the heatsink with even pressure on opposing-corner screws. Run 2 hours, expect 5-12 C drop. Highest-impact fix on units >24 months old.

Replace the thermal pad between the heat-spreader and the heatsink (if model uses pads, not paste). Some BOX-series units ship with a thermal pad rather than paste at the heat-spreader interface. Measure the existing pad thickness with calipers (typically 1.0-1.5 mm). Source a same-thickness pad with >=7 W/mK rating - Thermalright Odyssey or Gelid GP-Extreme are reliable choices. Cut to size, replace, reassemble. Pad pump-out is rare but real on units cycled through cold/hot startup hundreds of times.

Add an external 120 mm push fan in front of the intake grille on a 12 V adapter. Strap or 3D-print a bracket that holds a high-CFM 120 mm server fan 5-10 cm in front of the miner's intake, on its own 12 V power adapter (do not steal current from the miner's PSU). Increases mass airflow over the heatsink by 30-60%. Common Mining Hacker mod for HS5 in summer climates. Drops chip temp 4-7 C and lets the internal fan run at lower duty - quieter overall.

Flash the latest stock Goldshell firmware via find.goldshell.com Ethernet upgrade path if you skipped Step 5. Some recent BOX-series firmware revisions have lowered the temperature target and improved the fan PWM curve. Ethernet only, never WiFi. If the upgrade fails mid-flash, you are now in the bricked state covered by the Goldshell - Firmware Bricked (Red & Green LEDs Stuck) page and you'll need a burn-*.img recovery from hello@goldshell.com. Have an SD card and adapter ready before you start.

Rule out a marginal hashboard by swap-testing. If you have two same-model BOX-series miners, swap the hashboards between chassis (note all connector orientations, take photos). If the high temperature follows the hashboard to the new chassis, the hashboard is the source - usually one chip running 8-15 C hotter than its peers due to silicon-lottery variance or solder fatigue. If the high temperature stays in the original chassis, it's the cooling stack (heatsink, fan, paste) that's degraded. This isolation is 30 minutes of work and saves hours of guessing.

Stop and ship to D-Central if any of these conditions are true: chip temp still >78 C after Tier-3 repaste, fan swap, clean intake, stock firmware, <=25 C ambient; per-chip temperature spread >10 C across the hashboard; heatsink base shows visible discoloration, warping, or solder reflow marks; hashrate is now <85% of nameplate even with temperatures back in spec - silicon damage already done; or you are not comfortable removing a hashboard or applying thermal paste. Book via the D-Central ASIC Repair page. Diagnostic flat fee CAD $75-150; component-level repair $120-300; full hashboard replacement $400-700.