IceRiver KS Hashboard Tester Fixture: Using It Yourself

Source the PSU. Decide model coverage first. KS0 / KS1 / KS2 / KS3L: a consumer-grade 750 W 80+ Gold ATX PSU is sufficient. KS3 / KS3M: server PSU with breakout (DPS-1200FB, DPS-2000BB, or refurbished APW9 brick). Buy used from local recycling or eBay; budget CAD $40-$120. Inspect for capacitor bulging before buying. New ATX 750 W units run CAD $90-$140.

Source the controller board. Raspberry Pi 4 (4 GB model recommended) CAD $70-$110. A spare Pi 3B+ works but is slower for serial logging. SD card 32 GB Class 10 CAD $12. Pi 5 V USB-C PSU CAD $15. A spare laptop also works (skip the Pi entirely) — you'll need a USB-to-TTL adapter for the serial side.

Source the serial / breakout parts. USB-to-TTL adapter (CP2102 or FT232RL) CAD $8-$15. Inline 60 A ATC blade fuse + holder CAD $8. E-stop mushroom switch (rated >=10 A 12 V DC) CAD $18. Banana jacks for breakout (4 pcs) CAD $10. Heavy-gauge silicone wire (10 AWG for 12 V rail, 18 AWG for signal) CAD $20. XT60 or Anderson Powerpole connectors for the hashboard input CAD $12.

Source the IceRiver wiring harness. This is the model-specific part. Each KS variant has a different controller-to-hashboard connector. Easiest path: buy a salvaged controller-to-hashboard cable from an IceRiver parts seller (Zeus, AliExpress) and adapt — CAD $8-$25 per cable type. Do not try to recreate the connector pin-out from scratch on your first build.

Lay out the bench. Plywood baseboard 45 cm x 60 cm (CAD $15). Mount: PSU at the back, breakout panel in the middle, Pi on a M2.5 standoff in the front-right corner, hashboard cradle on the left. Cable routing: 12 V run is the shortest possible straight line from PSU to hashboard — never a bundle, never under signal cables.

Wire the PSU. ATX: short the green PS_ON to any black GND so the PSU comes on without a motherboard. Use the yellow +12 V and black GND rails — combine 2-3 yellows in parallel for KS3-class current. Server-PSU: install the breakout board per its included guide; jumper its enable pin. Test with a no-load multimeter check: 12.0 V +/- 0.2 V expected.

Install the inline fuse. 60 A ATC blade in series with the +12 V going to the hashboard input. Solder + heat-shrink. Crimp lugs are acceptable but solder + heat-shrink is more reliable for bench use. Don't use household twist-on connectors.

Install the E-stop. Wire the E-stop in series with the PSU's PS_ON-to-GND jumper from Step 6. Pressing the mushroom kills the 12 V rail. Test it: power up no-load, slap the E-stop, multimeter goes to 0 V within ~200 ms. If it doesn't, fix the E-stop wiring before any board touches the bench.

Mount the breakout panel. Banana jacks for +12 V (red), GND (black), TP872 (yellow), TP873 (white). Label each jack with a P-touch label. The panel is for measurement only — never carry hashboard load through banana jacks.

Cold continuity sweep. Power off, multimeter on continuity. Verify: PSU 12 V -> fuse -> E-stop -> hashboard input is continuous. PSU GND is continuous to hashboard GND. Banana-jack +12 V is continuous to PSU +12 V. No +12 V to chassis or GND. If any continuity check fails, fix wiring before applying power.

Flash the Pi base OS. Raspberry Pi OS Lite 64-bit, SSH enabled, fixed IP on a wired connection (don't bench-test over Wi-Fi). Update apt, install python3-pyserial, screen, git. Do not install desktop bloat.

Build the IceRiver controller-emulator stack. Clone D-Central's bench tools or build from public IceRiver firmware sources — extract just the hashboard-driver portion and the chip-enumeration logic. Strip the network / UI / pool / API layers. What remains is ~3-5 MB of code that pokes the hashboard's serial bus and reads nonces. Do not run a real pool from this Pi — bench, not production.

Wire the serial console. USB-to-TTL adapter TX -> hashboard debug RX, adapter RX -> hashboard debug TX, adapter GND -> hashboard GND. Never connect adapter VCC to anything (hashboard self-powers off the 12 V rail). On the Pi: `screen /dev/ttyUSB0 115200` shows the live serial output.

First-board calibration. Connect a known-good hashboard. Power up the 12 V rail, then start the Pi controller-emulator. Watch the serial console: chip-enumeration messages, then chip-status reports. Confirm the fixture sees the expected chip count for the model (4 for KS0, 12 for KS2, 18 for KS1, 56 for KS3L, 112 for KS3, etc.). If the fixture sees fewer chips than nameplate on a known-good board, your wiring or firmware blob is wrong, not the board.

Calibrate the burn-in routine. Run --burn-in --duration=24h on the known-good board. Log to a CSV: timestamp, chip count, hashrate, PSU voltage (read separately with the multimeter on a per-hour cadence, or — fancier — via the Pi's ADS1115 ADC if you wire one in). At end, the calibration baseline is the report you compare future repairs against.

Stop DIY when: you need to test KS5L / KS5M (multi-board topology, the fixture above doesn't natively support it without a much larger PSU and a multi-board harness — ship those to D-Central until you've built a v2 fixture). Or when you need to test more than ~3 boards in parallel (single-board fixture is the sweet spot; multi-bay fixtures are a different engineering problem). Or when the customer is paying for a documented test report — at that point a salaried tech with a Zeus-grade fixture is faster than your time.

D-Central's bench process. D-Central runs a pair of fixtures — one for KS0/KS1/KS2/KS3L (lower-power), one for KS3/KS5L/KS5M (server-PSU class). Each board: cold continuity, no-load voltage, controller bring-up, 1 h thermal-load run, cold-cluster IR sweep, 24 h burn-in if the customer paid for it. Pass record goes back to the customer with the board.

Document & ship. When you ship a repaired board back: include your bench's pass record (chip count, hashrate, PSU voltage stability over the burn-in window, before/after thermal images). Customers don't get this from anywhere else in the IceRiver retail world. Charge for it; it's a real differentiator.