Goldshell – Ethernet Link Down / No Network Detected

Power-cycle the miner properly. Pull the AC plug at the wall for a full `60` seconds, then plug back in. Wait `90` seconds for boot to complete and the network stack to bring `eth0` up. Watch the Ethernet port LEDs as the miner finishes booting. Many `ETH_DOWN` calls clear themselves on a clean cold boot — especially after a thunderstorm or brownout that wedged the PHY's PLL state. If link returns, you're done; the PHY just needed a hard reset. If it doesn't, continue.

Swap the Ethernet cable with a known-good `<5`-year-old CAT6 patch. Don't trust any cable that's been in the wall for more than five years, has visible kinks, has been stepped on, or has run near AC mains without shielding. Plug both ends firmly until you hear the click. If your network LED comes up immediately, the cable was the entire problem and you've saved an afternoon. Keep the bad cable as evidence and replace it with a properly-rated CAT6 run.

Swap the switch port. Move the cable to a different port on the same switch / router. Wait `60` seconds. If the LED lights, your original switch port was failing — re-flash the switch firmware or replace it. Bonus: try a brand-new dumb gigabit switch (`$20-30` consumer unit from any electronics shop) between the router and miner to rule out the switch entirely. Document which port failed for future reference.

Plug the miner directly into the router's LAN port. Bypass the entire switch chain and plug the miner straight into the gateway router. If the link LED lights here, you have a switch / cabling problem in the rest of your plant — not a miner problem. This is the cheapest and fastest way to confirm the miner Ethernet is healthy without opening the chassis.

Visually inspect the Ethernet jack on the miner. From the back, look at the RJ45 jack closely under good light. Bent retention clip, recessed metal pins, brown scorch mark on the chassis vent above the jack — any of those are physical damage and you're in Tier 3 / 4 territory regardless of what the cable test says. A scorched magnetics module needs a hardware repair, not a network reconfig.

Direct-connect the miner to a laptop. One Ethernet cable end-to-end, miner to laptop, no switch in between. Manually set the laptop's Ethernet adapter to IP `192.168.1.100`, mask `255.255.255.0`, no gateway, no DNS. Power-cycle the miner, wait `90` seconds, and try `ping 192.168.1.1`. Modern Goldshell firmware self-assigns `192.168.1.1` if DHCP fails. A successful ping confirms the miner Ethernet is alive — your LAN was the problem. No ping but link LED on = continue with forced speed/duplex.

Force-negotiate `10 Mbit half` on the laptop NIC and retry. Goldshell BOX miners are `10/100 Mbit` only — they cannot negotiate gigabit. A `10/100/1000` switch or NIC with broken auto-neg can fail to find common ground with a `10/100`-only PHY. On the laptop NIC properties, manually set speed/duplex to `10 Mbit half-duplex`, replug the cable, and try the ping again. If it works at forced speed but not at auto-neg, the PHY's auto-neg state machine is dying — Tier 3 / 4.

Multimeter the control-board power rails (chassis open). Open the two Phillips screws on the bottom of the BOX chassis. With the miner powered, probe the `3.3 V` rail (look for the voltage callout on the silkscreen near the PHY) and the `12 V` input. Expect `3.30 V ± 0.05` and `11.8-12.2 V` sustained. Sagging or oscillating `3.3 V` = on-board regulator or output cap is dying, and it's starving the PHY. Replace the regulator + cap, or escalate to control-board swap.

Re-seat the Ethernet jack solder if you spot a cold joint. Under good light and magnification, look at the underside solder pads on the RJ45 jack. Cracked or fractured solder around the jack pins is a known failure mode after years of cable insertion / removal stress. A clean reflow with flux and a fine-tip iron at `360 °C` for `2-3` seconds per pin restores the joint without removing the jack. Don't attempt this if you've never reflowed through-hole solder before — escalate to Tier 3.

Wireshark a boot capture from the laptop direct-connect. With the laptop direct-connected (Step 6) and Wireshark running on the laptop NIC, power-cycle the miner. Capture the first `120` seconds. Any frame from a `18:2A:7B` / `5C:CF:7F` / `24:C9:A1` source MAC = PHY is at least partially alive — chase auto-neg or magnetics. Zero frames for `120+` seconds = PHY is silent and the chip is the suspect. Save the .pcap file for the bench if you ship the board.

Serial console into the control board. Open the chassis fully. On BOX-series, find the 4-pin `GND/VCC/TX/RX` UART header on the control board near the Ethernet jack — pinout on silkscreen for most rev-A and rev-B boards but moved on later rev-C runs (probe with a multimeter to confirm `GND` if unsure). Connect a CH340 / CP2102 / FT232 USB-TTL adapter at `115200 8N1`. Open PuTTY / minicom / `screen /dev/ttyUSB0 115200`. Power-cycle the miner. Capture the full boot log. On KD6 / LT6 / CK6 chassis, serial access is via the CM4 module's GPIO header — see the Raspberry Pi CM4 datasheet for pinout.

Inspect the boot log for PHY-init errors. In the captured serial log, search for `eth0`, `mii`, `phy`, or `MAC` lines. A clean boot shows the PHY initializing, link negotiating, and `eth0` entering `state UP`. PHY-init errors, MII bus timeouts, or `eth0` never reaching `UP` — even with a known-good cable on a known-good port — confirm the PHY is the failure. If the boot log never reaches a login prompt at all, the control board is booting badly for non-Ethernet reasons — branch to the control-board failure guide.

Reflow the PHY chip (advanced rework, BOX-series only). With the heatsink removed and the chassis on a preheat plate, heat the bottom side of the control board to `~150 °C`, then top-side hot-air at `300-320 °C` for `25-30` seconds directly over the PHY QFN package. Let it cool naturally — do not move the board for `10` minutes. A drifted-but-not-dead PHY can sometimes recover from this; a fully dead PHY won't. If the symptom returns within `30` days, the chip is dead and only a chip swap will fix it.

Replace the PHY chip (BOX-series, bench rework). Source the same PHY from a parts-grade donor board or from electronics distributors — the part on most Goldshell BOX-series is a Microchip / Realtek `10/100 PHY` in a `5x5 mm` QFN-`32` package; verify the exact silkscreen marking on your specific revision before ordering. Hot-air desolder at `330-350 °C`, clean the pads with flux + braid, place the new chip, reflow at `300-320 °C` for `30` seconds. Reassemble and re-test. This is real-bench rework — don't attempt without prior QFN reflow experience.

Replace the entire control board (BOX-series, swap kit). Faster and cheaper than chip-level repair if you can source a parts-graded donor: order a parts-grade BOX control board from D-Central or a community parts swap, transfer the hashboard / PSU / chassis, and re-test. The hashboards and PSU are model-specific but the control board is largely interchangeable across BOX-family hardware revisions of the same generation. Verify firmware compatibility before the swap — wrong firmware version can soft-brick the new board.

Stop DIY and book a D-Central repair slot. Stop when (a) you've confirmed via Wireshark + serial-console boot log that the PHY is silent and the cable / port / switch chain is innocent, (b) the magnetics module is visibly damaged or the RJ45 jack is mechanically destroyed, (c) you've reflowed the PHY once and the symptom returned within `30` days, or (d) the on-board `3.3 V` rail is sagging and you don't have the bench equipment to chase the regulator failure. Book ASIC Repair: https://d-central.tech/services/asic-repair/ with a note that link is dead and the board has been bench-isolated.

Ship to D-Central. Pack the miner with anti-static protection on the control-board side, double-box with `5 cm+` of foam padding on every side. Include the PSU + the Ethernet cable you've been using + a printed note with observed symptoms, model, serial, MAC (if known from prior history), router DHCP screenshot if available, and contact info. The MAC and DHCP screenshot save the bench `30-60` minutes of re-diagnosis. Canada-wide shipping standard; US / international welcomed — declare as `used computer equipment for repair, value $X CAD` to keep brokerage reasonable.