Goldshell KD-BOX II / AL-BOX II – New Hardware Issues

Power-cycle the miner properly. Unplug 12 V from the barrel jack. Wait 30 full seconds — not 5, not 15, a full 30 — to drain the bulk caps and let the eMMC commit any pending writes. Re-plug. Wait 3 minutes for full boot. A surprising fraction of 'broken' KD Box II tickets are mid-boot stalls that resolve cleanly with a real power-cycle. If symptoms persist, proceed to the next steps.

Audit the chip count and hashrate baseline. Log in at `http://<miner-IP>`. Navigate to `Status`. Record `Total chips`, current hashrate (1m, 15m, 24h averages), per-chain temperatures, and firmware version. Save this baseline. Re-check weekly for the first 30 days. If chip count drops, you've caught a chip dropout early — RMA warranty window matters, document the date.

Roll back firmware over Ethernet if you're on a flagged version. Check the Goldshell GitHub firmware issues for KD Box II problems with your current version. If flagged, download the previous version from the Goldshell firmware portal, connect via Ethernet only, navigate to `System → Firmware → Upgrade`, select the older `.img`, click Upgrade. Do NOT use WiFi. Wait 5-10 minutes for upgrade and reboot cycle.

Change the default SSH and web UI password immediately. SSH or web UI: change `admin` / `123456789` to something strong. Goldshell default credentials are actively targeted by botnets. Even on a LAN-only deployment, default credentials are a liability — guest WiFi cross-pollination, compromised IoT devices on the same subnet, and lateral movement from a phished workstation all happen.

Disable WiFi entirely once Ethernet is working. `System → Network → WiFi → Disable`. The KD Box II's WiFi stack is the single biggest source of post-launch bricks because it interacts badly with firmware updates. Run on Ethernet permanently. If you genuinely need wireless, use a Wi-Fi-to-Ethernet bridge / travel router and let the miner see only Ethernet.

Verify the 12 V rail under load. DMM in DC mode on the 20 V range. Probe at the barrel jack (or open the case and probe at the control board input pins). Expected: `12.0 ± 0.3 V` while the miner is hashing flat-out. If the rail sags below `11.4 V`: the PSU brick is underspec'd or aging. Swap to a known-good `12 V / 5 A` brick. The KD Box II shipped PSU brick is the cheapest weak link in the kit.

Log wall voltage for 24 hours. A Klein Tools ET920 USB voltage logger or any cheap Kill-A-Watt logs the outlet. Run for 24 hours. If wall voltage dips below `108 V` (120 V circuits) or `225 V` (240 V circuits) during evening peak (6-10 PM), you have line-side sag. Move the miner to a different circuit, add a UPS, or accept the daily reboot. KD Box II is unusually sensitive — a `300 VA` UPS is enough at `~$80 CAD`.

SD-card recovery for a bricked unit. If both LEDs are stuck red+green after a failed firmware upgrade, firmware is corrupted. Email `hello@goldshell.com` requesting `burn-kd-box-ii.img`. Subject: `burn-kd-box-ii.img firmware recovery request - SN: <serial>`. Response is `24-72 hours`. While waiting, prepare an `8 GB` Samsung EVO Plus or SanDisk Industrial microSD with balenaEtcher. Once the image arrives, flash to SD, insert into miner's SD slot, power-cycle, wait 5-15 minutes for eMMC reflash.

Open the case and reseat the hashboard ribbon cable. Power off, unplug. Remove the 4 case screws. Identify the ribbon cable connecting hashboard to control board. Disconnect (note orientation — mark with a Sharpie if needed), inspect for bent pins, oxidation, or debris. Reconnect firmly until you hear/feel the click. Reassemble. Power on. Re-check chip count. A surprising number of 'missing chips' complaints on early KD Box II units are intermittent ribbon contact, not actual chip death.

Apply fresh thermal paste on the ASIC chips if hashrate is low and chip temps are high. KD Box II ships with a thin layer of factory paste that dries faster than the marketing implies — typical service life is `12-18 months` on a unit running 24/7 in a warm room. Disassemble, clean old paste with 99% isopropyl alcohol, apply a thin uniform layer of Arctic MX-6 or equivalent, reassemble. Service life of fresh paste: `18-24 months`.

Identify and replace dead chips at the BGA level. If chip count is consistently below nameplate after every diagnostic above, individual chips have died. Pull the hashboard, identify dead positions from the per-chip table, and reflow with a preheat-plus-hot-air station (preheat 150 °C, top-side 280-300 °C for 25-30 seconds — Kadena/Blake silicon tolerates a reflow cycle similarly to SHA-256 silicon but at slightly lower top-side temps). Reflow first; if chip is genuinely dead, replace with a salvaged chip from a parted-out KD Box II hashboard.

Inspect and replace bulk caps on the hashboard input. Bulging electrolytics or cracked MLCCs near the chip array are repair territory. Standard low-ESR replacements: Panasonic FR / FM or Nichicon PW series. Match the original capacitance and voltage rating; uprate to `25 V` if the original was `16 V` for headroom on a unit you intend to run for 5+ years.

Dump the eMMC and inspect for malware before reflashing if the miner was ever exposed to the public internet. Pull the control board, read the eMMC via USB-C reader / JTAG. Mount partitions on a Linux workstation. Inspect `/etc/init.d/`, `/usr/local/bin/`, `/tmp/` for tampered scripts, rogue binaries, or curl/wget calls to suspicious IPs. If you find tampering, follow the full forensic path before reflashing.

Direct eMMC write of the recovery image if SD-card recovery fails. With the eMMC dumped to a workstation: `dd if=burn-kd-box-ii.img of=/dev/sdX bs=4M status=progress conv=fsync`. Verify with `sha256sum`. This bypasses the SD-card bootloader entirely — use this when the bootloader itself is corrupted, not just the rootfs. Mismatch on `sha256sum` means a bad eMMC block; move to Tier 4.

24-hour burn-in at nameplate before declaring the repair done. Connect to your pool. Monitor hashrate, per-chip status, temperatures, and any reboots for 24 hours. Successful repair: nameplate hashrate `± 5%`, no chip dropouts, no unexplained reboots. Hashrate creeping low or chips dropping: deeper hardware problem, escalate to Tier 4.

When to stop DIY. Send to D-Central when: (a) more than 5 chips are dead on a hashboard, (b) the eMMC is physically damaged or write-verifies bad after 3 attempts, (c) the control board has compound failures (dead Ethernet PHY, dead fan connector, bad PMIC), (d) post-Tier-3 hashrate is still below 70% of nameplate after 24 h burn-in, or (e) you suspect malware tampering and want a clean-room reflash. Round-trip shipping is cheap on a unit this small; turnaround typically `5-10 business days` from receipt.

What D-Central does at the bench. Bench-validated `burn-kd-box-ii.img` library, programmable bench load for full input-window characterization, BGA rework station for chip replacement using parted-out KD Box II hashboards from the salvage inventory, eMMC reflash via direct JTAG when SD-card path fails, post-repair `24-hour burn-in` at nameplate. We're the only Canadian shop documenting Goldshell-specific repair procedures publicly — no overseas-shipping coin-flip, no 6-week mystery turnaround.

Ship safely. ESD bag around the control board. Foam-cradled rigid box for the whole miner. Include a note: model, serial, manufacture date from the rear sticker, firmware version, what symptom you saw, what you've already tried, and whether the device was ever port-forwarded to the internet. Saves diagnostic time and your repair cost. Ship to D-Central HQ — Canada-wide shipping, US/international welcomed.