IceRiver Power Drops to 150W-200W: Diagnosing the Cause

Hard power-cycle from the rear rocker. `60 second` disconnect, then power back up. Watch the boot self-test (~`2 minutes`), then log per-board hashrate and rig wattage for 15 minutes after warm-up. If the rig recovers to nameplate `1,500 W` / `3,400 W` on cold start but drops back over hours, the fault is thermal — silicon and pads getting up to operating temperature and something giving way. That points squarely at thermal-interface degradation (Tier 3 Step 11). If the rig stays at `150 W-200 W` immediately after cold boot, the fault is hardware-locked or firmware-locked and you need to escalate to Tier 2.

Read per-board hashrate from the IceRiver web UI. Single most important diagnostic on a `150 W-200 W` event, and most operators never click into it. Navigate the dashboard to the per-chain status pane. Each board should sit at `~3.3 TH/s` on KS3M, `~4 TH/s` on KS5L, or matching the per-board math for your specific SKU. Record all three. If one or more reads `0 TH/s`, you've isolated a board offline. If all three read `0`, the cascade is system-wide. This step costs nothing, takes one minute, and tells you which branch of the diagnostic tree to walk.

Read `Temp1` and `Temp2` per board. Same dashboard, same minute. Healthy: `Temp1 = 45-55 C`, `Temp2 = 50-60 C` per IceRiver's official guidance. Any board at `Temp2 > 75 C` with fans pinned is the thermal governor pulling that chain back to idle and dragging rig wattage to the floor. Per-board asymmetry of `>= 5 C` between siblings is a smoking gun for thermal-interface failure on the hot board specifically — fix that pad and the governor releases.

Verify ambient at the intake with an IR thermometer at the front grille - not room-middle. KS-series specs `0-35 C` operating envelope; the practical sweet spot for sustained nameplate hashing is `<= 30 C` ambient. Above `30 C` and you're in heat-throttle territory before any other fault enters. If you're chasing a summer wattage drop and your basement is at `32 C`, fix the ambient before tearing into the chassis. Window A/C, summer duct, or relocate the miner. Canadian basements in February run a KS5L at nameplate happily; a Texas garage in August won't.

Clean both fans and the front grille filter. Canned air, short bursts, hold each `12038` fan impeller still with a plastic probe to avoid back-spin EMF on the bearing. Blow out heatsink fins downstream of the fans. Wipe the front grille filter (if installed). Pet hair, basement carpet drift, garage workshop dust - any of these halves effective airflow inside `6 months` and the controller silently downclocks chains to compensate, dropping rig wattage to the `150 W-200 W` floor. Single Tier 1 step that recovers a meaningful fraction of cases without disassembly.

Roll firmware back to last-known-good. If wattage cliff appeared right after a firmware update, this is the suspect. Pull prior KS-series firmware from https://www.iceriver.io/firmware-download/ - match the variant exactly. KS3M firmware on a KS5L bricks the controller; KS5M firmware on a KS5L mis-clocks the boards; wrong variant is a brick risk. Flash through the dashboard updater (or SD-card recovery if dashboard is unstable), wait for reboot, observe per-board hashrate and rig wattage for 30 minutes after warm-up. If the rig returns to nameplate, firmware regression confirmed - pin to that build. Verify-flag: community reports of regression-prone 2024-2025 KS5L / KS3M builds; specific version numbers pending D-Central bench cataloguing.

Probe the PSU output rail under full load. Multimeter on DC volts. Locate the PSU-to-controller / PSU-to-board power connectors. Probe `V+` to `GND` while the miner is trying to hash at full load - not at idle, not at boot. KS-series PSU expects `12 V` main rail with `11.8-12.2 V` sustained tolerance under continuous nameplate draw (verify-flag against your specific PSU revision and hashboard count). Below `11 V` sustained = PSU tired or circuit undersized. Above `12.4 V` = regulation drift, replace before damaging boards. `BP-H-3640` is the canonical IceRiver-spec replacement (verify-flag against current service catalog).

Verify line voltage at the wall outlet under load. Voltage logger on the outlet feeding the miner, log for 30+ minutes during attempted full-load mining. North American `120 V`: expect `115-125 V`. `240 V` split-phase: expect `235-245 V`. Line voltage sag is the single most-diagnosed mystery low-power cause across the entire ASIC repair industry. Evening sag during neighbourhood peak (`6-10 PM`) tracks A/C and dryer cycles in North American residential. If your KS5L wattage drops evening and recovers overnight, fix upstream wiring - dedicated `240 V` circuit strongly preferred - before chasing any hardware fault.

Reseat all hashboard data and power connectors. Power off at the rear rocker, wait `60 s` for caps to bleed, remove the top cover. Each KS-series hashboard connects to the controller via a ribbon (data) and a heavier power connector. Disconnect each, visually inspect contacts for blackening, corrosion, bent pins, or oxidation, reseat firmly until you feel the positive click. Apply a trace of dielectric grease on the power connectors to immune them against chassis vibration backing them out over `12+ months`. Partially-seated power connector causes voltage droop on that board only - exact symptom: one board offline, rig wattage drops to `150 W-200 W` because the cascade pulls remaining boards to idle.

Slot-swap diagnostic. Power off. Label the three KS-series hashboard slots `0/1/2` with tape. Move the underperforming (zero-hashrate) board to a known-good slot, move a known-good board into the suspect slot. Power on, log per-board hashrate after `5 minute` warm-up. If the fault follows the board (previously-bad board still reads zero in its new slot), the board is the problem - Tier 3 thermal service or Tier 4 component-level repair. If the fault stays in the slot (a different board now reads zero), the controller-side path (cable, connector, voltage rail to that slot) is the problem - Tier 4 controller swap. Most decisive test in the whole tree, four minutes, zero parts.

Re-pad / re-paste the underperforming board's thermal interface. High-leverage Tier 3 fix for the ~35% of `150 W-200 W` cases driven by thermal-interface degradation cascading into the firmware governor. Power off, remove the top cover, lift the heatsink off the affected board (count and remove heatsink retention screws - KS5L typically `8`, KS5M `10`, KS3M typically `6`; verify-flag). Carefully separate heatsink from the `1004LV100` chip array. Clean residual thermal compound from chip dies and heatsink with isopropyl `99%` and lint-free wipes. Apply fresh `0.5 mm` graphite/silicone thermal pads (verify-flag - mount style varies by revision; some units ship paste-only), or apply Arctic MX-6 / Thermal Grizzly Kryonaut paste in a thin even layer across each die. Reassemble with even torque on retention screws - uneven mount pressure re-creates the same fault. Power on, observe per-board hashrate, rig wattage, and `Temp2` for 30 minutes.

Inspect the `1004LV100` chip array for visible damage on the offline board. While the heatsink is off, inspect each chip with a magnifier. Look for: cracked package edges, lifted BGA corners, scorched solder mask near a chip, discolouration on the PCB substrate around a chip, or a chip visibly shifted from its footprint. Any of these = the chip is dead or dying, no thermal service will fix it. Document position of any damaged chip. Recovery path is either chip-level rework (preheat + hot-air at `310-330 C`, lift and replace with salvaged-grade `1004LV100` from D-Central parts inventory; verify-flag - IceRiver chip salvage market is thin compared to Bitmain BM1398) or full board swap from D-Central KS-series inventory at the bench.

Voltage-domain probe on the offline board. Power off, multimeter on DC volts. Each KS-series hashboard has a local voltage regulator chain converting the `12 V` main rail into `1004LV100`-required core voltage (verify-flag - exact core voltage varies; IceRiver does not publish board schematics, D-Central pattern-matches against analogous Kaspa hardware). Probe the local rail under attempted load - sagging, pulsing, or completely absent (`0 V`) on an offline board = local PMIC / inductor / regulator drift on that specific board, which explains the chain going dark and the rig dropping to `150 W-200 W` standby. Hot-air rework job or full board swap. Territory between solo DIY and Tier 4 - parts ID requires either an IceRiver schematic (not published) or a salvaged reference board (D-Central has one). Most operators escalate to Tier 4 here.

SD-card firmware recovery flash for sensor-subsystem fault. If `Temp1`/`Temp2` are reading garbage values (e.g. `-10 C`-style impossible readings) or boards are flagged offline by the controller despite passing slot-swap, the controller's sensor subsystem may be faulted and firmware is flying blind on thermal - manifests as cascading offline boards and rig wattage at the floor. SD-card recovery: pull the official KS-series image for your exact variant from https://www.iceriver.io/firmware-download/, write to a microSD with BalenaEtcher, insert into the controller's SD slot, power-cycle, watch for the recovery LED pattern (verify-flag - recovery LED behaviour documented per-variant in IceRiver KB). Wrong image variant bricks the controller permanently - confirm exact SKU before flashing. DCENT_OS is NOT applicable here - IceRiver runs `1004LV100` Kaspa silicon, completely different architecture than DCENT_OS targets. Stick with IceRiver's official firmware archive.

Stop DIY when any of these are true: visible heat damage, blistered solder mask, or burnt-component odor on any board or controller; a second board on the same rig drops offline within `30 days` of a first repair; suspected `1004LV100` chip-level damage on more than one chip per board; local voltage-domain regulator drift confirmed by Tier 3 probe on multiple boards; thermal service attempted and one or more boards still won't return to nameplate hashrate; rig wattage cliff persists after a clean firmware reflash and per-board confirmation that boards are physically online; you attempted SMD rework and lifted a pad; miner ran for hours with `Temp2 > 85 C`. Test-fixture territory. Book a D-Central ASIC Repair slot at https://d-central.tech/services/asic-repair/. Western English-language IceRiver repair authority - Zeus Mining (China) is the only competitor and their trust speaks for itself.

What D-Central does at the bench for KS-series `150 W-200 W` faults. Diagnosis against a known-good KS-series reference rig with calibrated PSU and matched ambient - separates silicon-side from environment / supply faults inside an hour. Per-board hashrate and per-board power-draw isolation with a KS-series hashboard tester. `1004LV100` chip-level rework when a single failed chip is the cause - preheat plus hot-air, salvaged-grade or new-old-stock chip placement, full reflow and reseal. Full thermal service - paste / pad refresh on all boards, controller reseat, fan inspection. Voltage-domain regulator replacement on the affected board if PMIC / local regulator drift is the cause. Controller swap from D-Central KS-series inventory if controller-side fault is diagnosis. PSU swap to known-good `BP-H-3640` (verify-flag) if PSU is the cause. 24-hour nameplate burn-in at the unit's expected wattage with both fans monitored, per-board hashrate logged, and rig wattage logged before unit ships back. Verify-flag: nameplate values per SKU directional - confirm against exact unit revision.

Ship the whole miner, not the boards alone. KS-series hashboards are fiddly to pack safely separated from the chassis - the `1004LV100` chips are cornerless and the heatsink-board sandwich can shift in transit. Double-box the full unit, optionally separate the boards into anti-static bags inside if you must ship hashboards-only. Include a printed note with the dashboard wattage and per-board hashrate readouts, current firmware version, observed `Temp1`/`Temp2` per board, ambient at the install site, circuit voltage, and any error codes. That note saves D-Central diagnostic hours, which saves you repair dollars. Canadian customers ship to our Quebec bench, miner back in `5-10 business days`. US and international welcomed - repair desk handles international shipping paperwork.