IceRiver Chip Count Mismatch (9, 26, 52): Partial Hashboard Failure

Hard power-cycle the miner at the breaker for 30 seconds — not a soft reboot. The IceRiver miner daemon can wedge after a firmware update, a brownout, or a transient I2C error during chip enumeration. A clean cold start lets the enumeration sweep run from a known boot path and clears any state that's pinning the count to a partial value. About 12 % of `chip count mismatch` reports we see at the bench resolve at this step alone, before any hardware work.

Wait the full 5 minutes after boot before reading the chip count. Enumeration on KS hardware completes in 2-3 minutes typically, but cold hardware in a cold basement can take longer. Reading the count too early gives a false 'partial' on a chain that's still enumerating. Watch the count climb board-by-board — that is normal, not a fault. Declare the count stable only after a 5-minute window with no further change.

Power off at the breaker. Open the chassis. Visually inspect every ribbon and `12 V` lead for partial seating, blackening, oxidation, bent pins, or shipping damage. A visibly disturbed connector is almost always your fix — vibration walks pins out of seat over time, and a partially mated contact passes idle traffic but loses one or two chip enumerations under timing pressure. That's exactly the signature of `8/9` or `25/26`.

Verify firmware version against the chassis revision label. IceRiver shipped multiple silicon revisions of `KS0 Pro`, `KS5L`, and others — and OTA images are not cross-compatible. Mismatched firmware can enumerate most of a chain and miss one or two chips at the boundary. Cross-reference your installed firmware against the IceRiver download page before any hardware work — this is the cheapest possible fix.

Factory reset the miner (20-second button hold, red status LED flashing confirms) and reflash the matching firmware for your specific hardware revision. About 5 % of `stuck partial count` reports we see at the bench resolve at this step. Do not flash 'newer' firmware as a fix unless you've confirmed it matches your revision — wrong-revision newer firmware makes the problem worse, not better.

Cold-chip thermal scan with an IR thermometer or thermal camera (FLIR ONE Pro is sufficient). With the chassis open and the miner running at idle post-enumeration, scan the suspect chain. Healthy chips run `45-65 °C` at idle. The dead chip reads cold — ambient or near-ambient — because it dissipates no power. Adjacent chips past the dead one on the daisy-chain may also read cooler. Note the cold position and record it.

Multimeter on DC. Probe `vCore` at the cold chip's local decoupling cap (refer to KS hashboard reference photos at zeusbtc.com if you don't have a board-specific schematic). Expect `0.4`-`0.7 V` sustained. Reboot the miner and watch the meter during the enumeration window (first 60-90 seconds). A clean rail under load = chip is dead → chip replacement. A sagging or oscillating rail = PMIC / cap fault on the chip's domain → component-level repair.

Reseat the suspect board between slots on multi-board models. Label slots `0` / `1` / `2` with masking tape. Move the failing board to a known-good slot. Reboot, recheck the chip count. Fault follows the board → bad board (chip / trace / PMIC). Fault stays in the slot → controller-side data lane (rarer; escalate to Tier 3 / 4 bench work).

Reseat the chain ribbon with rotation. Disconnect the chain ribbon, re-orient the contact face by 180 °, reconnect. Sometimes a partially oxidized contact will mate cleanly on a fresh face. Reboot, recheck. About 8 % of `8/9` or `25/26` reports clear after a careful re-seat with re-orientation when a straight reseat didn't work.

Inspect the C19 power inlet and the AC cord on `KS3` / `KS5` / `KS5L` series. Discolored or scorched pins on the C19 plug can drop AC supply enough to make the PSU sag during the enumeration current ramp — the symptom surfaces as a partial chip count on the board with the highest current draw. Replace the cord with a 16 A-rated unit and consider replacing the chassis-side socket if user-accessible.

Component-level: identify and replace cracked MLCCs. Under microscope, scan the suspect chain's chip positions for hairline-cracked MLCCs in the `vCore` decoupling. Cracked caps are the #1 partial-count cause we see at the bench after cable / firmware. Replacement is `0402` or `0603` MLCCs in matching value, hot-air or fine-tip iron, careful flux work. Inexpensive parts; the time investment is the cost.

Component-level: trace repair on a broken serial-bus path. If scope shows the data line going silent at a specific chip position, the trace into that chip's serial-in is likely broken from PCB stress (shipping damage or chassis flex). Microscope, fine wire (`0.1 mm` enameled), flux, careful soldering. Validate with a continuity check before reassembly. Tier 3 minor surgery — slow and steady.

Reflow the cold chip's BGA. If the chip is dead but the package itself is intact (no scorching, no swollen pads), a reflow can restore the chip-to-board connection — kHeavyHash chips on KS hardware tolerate a reflow cycle reasonably well. Preheat to `150 °C`, top-side hot air at `300-320 °C` for 30 seconds. Re-paste, reassemble. Outcome: 30-50 % chance the reflow restores enumeration depending on whether the underlying fault was a cracked solder ball (yes) or dead silicon (no).

Inspect and replace the chain's PMIC if the rail sags under load. PMIC replacement requires hot-air rework + microscope + the matching part number. PMICs on KS hashboards are not always documented publicly — Zeus Mining has the most complete parts cross-reference. Source the matching PMIC, rework, retest under instrumented load. Validate the rail is back inside `0.4`-`0.7 V` sustained during enumeration before reassembly.

Re-paste the cold chip's heatsink contact. Rare but real: a heatsink that has lifted off thermal contact at one chip position can let that chip drift thermally past its enumeration timing margin without killing the chip outright. Clean both surfaces with `99 %` IPA and lint-free wipes, apply fresh paste (Arctic MX-6 or Kryonaut), reassemble with proper torque. Recheck the chip count after a clean cold boot.

Stop DIY when: chip replacement is required and you don't have hot-air rework, microscope, and matching kHeavyHash silicon stocked; PMIC replacement requires the part number cross-reference and you don't have it; trace repair is on an inner PCB layer (X-ray territory). Ship to D-Central. Honest economics: on a `KS0` running `8/9`, the repair may exceed the used-market value of a clean KS0 — we'll diagnose, quote, and let you decide.

D-Central bench process: per-chain isolation under instrumented load, scope-level data-line capture during enumeration, thermal imaging across all chip positions, MLCC / PMIC component-level repair, BGA reflow on candidate chips, salvage-grade chip swap when inventory is available, post-repair 24-hour burn-in at nameplate hashrate. Turnaround `7-14` business days from receipt. Canada-wide shipping; US and international welcomed.

Ship safely. Hashboards in anti-static bags. Whole chassis double-boxed with at least 5 cm foam every side. Include a note: model (`KS0` / `KS3M` / `KS5L` etc.), hardware revision label, observed chip count (`8/9` / `25/26` / `51/52`), what you tried, contact info. The fewer guesses we have to make at the bench, the lower your repair cost — saves diagnostic time, which saves you money.