Extras — optional hardware and practices#

Additions that are not part of the core migration but earned their place on a real machine.

USB WiFi dongle (when the onboard radio dies)#

The onboard RTL8189FTV is a known single point of failure — on this unit it died of PSU-inflicted causes (mmc: error -110 whilst initialising SDIO card on every boot, on stock firmware too, through ~21 attempts across two kernel stacks; software revival — power-cycling the chip via its GPIO, longer post-power-on-delay-ms, lower SDIO clock — changes nothing). If you see that signature, stop debugging and plug a dongle into one of the board’s USB-A ports.

Pick an MT7601U stick (sold as LP-UW06 among others, lsusb ID 148f:7601): the mt7601u driver is in the Armbian kernel, the firmware blob ships in the image, there is no USB mode-switch dance and no competing out-of-tree driver — it is the only truly plug-and-play option. RTL8811CU-class sticks work too but bring usb_modeswitch plus two rival drivers.

Two files make it drop-in (bake them at preseed time, see clean-slate):

# /etc/systemd/network/10-wlan0.link — stable name off the driver
[Match]
Driver=mt7601u

[Link]
Name=wlan0

and the usual netplan wifis: wlan0: block (key-management: "psk", optional: true). With both in place the dongle associates on first boot with zero manual steps.

Stock PSU — keep it grounded, and a bolt-on replacement#

The stock 24 V supply fails often enough to plan for, and running it without a mains earth is what kills it — on one machine a dying PSU took the mainboard’s onboard WiFi chip with it (the failure behind the dead radio above). Always run it grounded.

A failing supply often shows itself first as the mainboard-cooling fan running flat out from cold — that’s voltage sag, so if the board fan screams from day one, suspect the PSU, not the fan. It can also present the other way round: peripherals — an MCU, Ethernet, the radio — dropping off one at a time while the board still boots is the supply on its way out, not a set of separate mainboard faults.

For a drop-in replacement that needs no fabrication, a MEAN WELL LRS-150-24 (24 V, 150 W) is bolt-on: two screws in the lower bay, move the cables across, done. It’s a more reputable supply than the stock unit, so likely more tolerant of abuse. (A 48 V supply is a separate speed mod, not a replacement.)

Extruder gearbox — grease it, steel gears if you run it hard#

The planetary reduction in the extruder is plastic — the brass sheen is grease, not metal — and it ships nearly dry. Over long runs the ring, satellites, and drive gear wear each other down; a worn set feeds unevenly and the extruder grinds filament instead of clicking. Two things help:

Re-grease before the first run with Molykote 30L / EM-50L. Lithium greases (Litol, Ciatim) darken and wet plastic poorly — don’t use them.
When it’s worn, go steel. Orbiter 2.0 / 2.5 hardened gears drop in. For the sun/drive gear specifically, one long-running report (three machines, weeks of 24/7) uses an “XHGMAK 0.5-module, 10–22 T, steel 45” gear bonded with Loctite 638 — it sits at ~0.02 mm clearance and won’t hold torque unbonded.

Toolhead board caps the chamber temperature#

The toolhead MCU is a commercial STM32F103, rated to 85 °C ambient. In a closed heated chamber (ABS/PA, ~60–65 °C air) the head board reaches 80–87 °C and can drop off the CAN bus, so on the Zero the practical chamber ceiling is set by the head electronics, not the hotend. If you need it hotter, people cool the head — pull the cover, add a small copper heatsink on the MCU and driver, or fit a 6010-fan duct; a higher-rated MCU board doesn’t help, since the same TMC2209 driver still overheats.

48 V speed mod#

The mainboard is built for 48 V — there’s a separate labelled 48 V input and two jumpers, factory-set to 24 V — but Sovol ships only a 24 V supply to hit a price point. Fit a 48 V PSU (a MeanWell UHP-350-48 sits flush in the under-bed bay) and move both jumpers; nothing in printer.cfg changes and run_current stays 3.5 A. It roughly doubles torque at high RPM — at 24 V the coils are starved by back-EMF and winding inductance — cutting a stock ~8-minute benchy to about 5. The 6 mm steel base plate needs drilling and tapping M3 for the new supply. This is a speed project, not the like-for-like replacement above.

BTT SFS V2.0 filament sensor#

The board’s three detect ports (silkscreen legend PLA_DET / TWINE_DET / DOOR_DET) share the same 3V3 / GND / signal pinout, which happens to match the SFS V2.0’s two split tails: the red-empty-blue tail carries VCC + the runout switch signal, the empty-black-green tail carries GND + the motion signal — the module powers up only with both plugged. Any two ports work; the signal pins by legend order are PB2 (PLA_DET), PE7 (TWINE_DET), PE8 (DOOR_DET). Cable colours vary between batches — trust the module PCB’s own port markings over wire colour.

Installing the SFS needs some disassembly, but only the chamber’s rear wall — that’s the whole teardown. The tempting mistake is to rip out the old stock sensor and its cable entirely while you’re in there. Don’t: that cable is routed the length of the printer (chamber → lower bay → door jamb), so a full removal means opening the lower bay too, cleaning the compound off its motherboard port, cutting every zip tie along the run, and reassembling it all in reverse. Instead, unclip only the old sensor from the cable end, zip-tie the loose end out of the way with the ties in the box, and leave the cable in its potted port. Feed the SFS’s own cable in through the chamber’s side window and land it on two adjacent free ports (PE7 + PE8 below).

Klipper config for green→PE7 (motion), blue→PE8 (runout), stock PB2 sensor removed from the path:

[filament_switch_sensor sfs_runout]
pause_on_runout: False
switch_pin: PE8
runout_gcode:
    PAUSE
    M117 SFS-S Empty
insert_gcode:
    M117 SFS-S Inserted

[filament_motion_sensor sfs_motion]
pause_on_runout: False
switch_pin: PE7
detection_length: 4.99
extruder: extruder
runout_gcode:
    PAUSE
    M117 SFS-M Static
insert_gcode:
    M117 SFS-M Moving

detection_length: 4.99 is deliberately forgiving — BTT’s 2.88 “high accuracy” value false-triggers on retract-heavy prints. Do not add [pause_resume] — mainsail.cfg already ships it and a duplicate section fails config parsing. If the stock PB2 sensor is physically out of the filament path, delete its config section too: a floating input is a false-runout generator. Verify polarity live with QUERY_FILAMENT_SENSOR with filament in and out; invert with ! on the pin if needed.

Backups that make experiments cheap#

Two layers, both proven in anger:

Full eMMC image (module in a USB reader): dd the whole device, keep the SHA256. This is the only backup that captures everything — partition table, boot partition, rootfs — and restoring it is one dd back. Take one before any risky campaign.
On-board tar pair for kernel experiments, taken while the system runs: tar czf /root/boot.tar.gz /boot and tar czf /root/modules.tar.gz /lib/modules/$(uname -r). Armbian kernel packages replace files in place (/boot is vfat — no symlinks, and the old /lib/modules/<ver> is deleted on upgrade), so a restore script that untars both plus a dead-man systemd unit (no network N minutes after boot → restore + reboot) turns a kernel test into a reversible operation even with no second access channel. A WiFi dongle as an independent path (see above) makes it fully safe.