Rumored IceRiver Capacitor Failures: What We Know (2026 Update)

Verify your unit's purchase date, model, and approximate shipping batch. Pull the invoice. Concentrate cap-failure diagnostic energy on KS5L units in the Q3 2023 - Q1 2024 early-production window failing inside <18 months, or any KS-series unit with >24 months of 24/7 operation regardless of batch. Units outside these cohorts are more likely chasing a different fault than the community-flagged batch issue. Don't open the PSU until you've at least matched cohort or eliminated AC sag - most 'cap failure' tickets at D-Central's bench resolve at Step 2 or Step 3 without ever opening anything.

Move the miner to a dedicated 240 V / 20 A circuit. NEMA 6-20 or NEMA 14-30. No shared loads, no extension cords, no power strips, no daisy-chains. Roughly half of 'KS5 cap failure' tickets resolve at this step before anyone opens the PSU. Unstable AC is the single biggest accelerator of cap aging - every volt of evening sag is more current through the bulk cap and more heat in the dielectric. If your panel doesn't have a free 240 V slot, this is a $300-$600 electrician job that pays for itself in cap longevity within 12 months.

Improve ambient cooling at the miner intake. IR thermometer at the front grille - not room-middle. Target <30 °C. Add ducting, summer A/C, ventilation, or relocate the miner to a basement / outbuilding with naturally cooler air. The Arrhenius rule says every 10 °C reduction in PSU enclosure temperature roughly doubles cap life. Single highest-ROI prevention measure after the dedicated circuit. KS5 spec allows 35 °C operating but every degree below that buys you months of cap lifetime.

Power-cycle from the rear rocker. Full disconnect for 60 seconds, then back up. Clears latched PSU protection state from a transient sag event. Not a soft reboot from the web UI - the controller needs a real cold start. If trips don't recur within 48 hours, you had a one-time grid event, not a hardware fault. If they recur, you have a real hardware or environmental issue and need to escalate the diagnostic.

Roll firmware to last-known-good if symptoms appeared after a recent IceRiver update. Pull the prior build from https://www.iceriver.io/firmware-download/ and match firmware variant exactly to your model - KS5L != KS5M != KS3M, wrong-model firmware bricks the controller. IceRiver has shipped builds with PSU-monitoring regressions that report false trip events. Rollback is 15 minutes. DCENT_OS is NOT applicable here - IceRiver runs on different silicon (1004LV100-class), DCENT_OS is Bitmain-Antminer-only.

Measure DC rail under load. Multimeter on DC volts, probe `V+` to `GND` at the PSU-to-controller harness while the miner is at full hashrate. Healthy: 11.8-12.2 V sustained. Sagging below 11.4 V confirms the PSU is the bottleneck but does not yet confirm caps vs MOSFETs vs transformer. Take readings every 10 minutes for an hour to capture thermal-soak behavior. This 5-minute test is the most decisive diagnostic in the whole tree - $5 of meter probe saves hundreds in misdirected parts.

Re-seat every PSU connector at the controller and at the hashboards. Power off, wait 60 s, kill AC at the wall. Disconnect, inspect for blackening / corrosion / bent pins / crushed shells, reseat firmly until you feel the click. Apply a trace of dielectric grease. Vibration over 12+ months of 24/7 operation walks connectors out one notch - surprisingly often, 'cap failure' tickets resolve at this step when the actual fault was a backed-out connector raising contact resistance and looking like cap-driven sag.

Whole-circuit resistance check. Power off, multimeter on resistance / continuity. Probe from the wall outlet's hot pin back to the breaker panel under no load - should read near-zero. Anything above 0.5 ohms indicates contact resistance somewhere - backstabbed receptacle, loose wire nut, tired breaker, or worn outlet contacts. A 0.5-ohm series resistance at 15 A drops 7.5 V and dissipates 113 W in your wall - fire hazard plus cap killer. Fix the wiring before blaming the caps.

Replace the wall outlet if grip is loose. A 15 A outlet that's been cycled 1,000+ times loses blade tension. Plug wiggles in the receptacle = high contact resistance under load = arcing on every load transient = stress on the PSU input that accelerates cap aging. A $5 commercial-grade NEMA 5-15 or NEMA 6-20 receptacle is 15 minutes of work. Always use the back-clamp or screw terminals - never the backstab push-in connections, they are the #1 cause of high-impedance outlet faults.

External visual inspection of caps through the chassis vents. Power off, AC unplugged, wait 5 minutes. Shine a flashlight through every chassis vent and PSU opening. Look for bulged tops, brown residue, leaked electrolyte, dome distortion, electrolyte smell. If you can confirm visible cap damage from the outside without opening anything, you have a real cap failure regardless of cohort - skip to Tier 3. If the caps look clean from outside and there's no smell, the PSU may not have a cap problem at all.

Open the PSU and document every cap. CRITICAL SAFETY: power off, unplug AC, wait 5 minutes, verify <10 V DC across the bulk cap with a multimeter before any tool touches the board. The bulk cap stores enough charge to kill you. Photograph the board before touching anything. Catalog every electrolytic: location, capacitance, voltage, ESR class, manufacturer, dimensions. KS5 BP-H-3640 family typically has one large primary bulk cap (~450 V / 470 uF), four to six secondary filter caps (~25 V / 1000 uF-class), and several small signal-rail caps. Verify-flag: silkscreen markings vary across PSU revisions; cross-reference photos against a known-good reference unit before ordering parts.

In-circuit ESR test every electrolytic. Bob Parker design ESR meter or equivalent ($30-$100). Reference values: 450 V / 470 uF bulk cap healthy <50 mOhm, failing >200 mOhm. 25 V / 1000 uF secondary filter healthy <100 mOhm, failing >500 mOhm. 16 V / 470 uF controller cap healthy <150 mOhm, failing >800 mOhm. Verify-flag: exact ESR thresholds vary by cap PN; these are directional reference values from D-Central's bench. While the tool is out, test caps that look fine - early-stage ESR climb is invisible to the eye, and Mining Hackers replace caps preemptively on a maintenance schedule.

Source replacement caps from reputable suppliers. Acceptable: Nichicon PW or PJ series; Rubycon ZLH, ZLJ, or YXJ; Panasonic FR or FM series. 105 °C-rated, low-ESR, Japanese-brand electrolytics. Match capacitance, voltage rating, ESR class, and physical dimensions exactly. Verify-flag: D-Central's bench keeps a stocking list of common KS5 PSU caps; contact the repair desk for a parts kit if you cannot source locally. Avoid: generic Chinese-brand caps from anonymous eBay sellers - counterfeits are common in this part class and they will fail inside a year regardless of how careful your install is.

Desolder the failed caps. Hot-air rework station at 350 °C for the smaller filter caps, or a temperature-controlled iron at 380 °C plus solder braid for individual leads. The bulk cap typically requires both hot air and an iron to lift cleanly without damaging the PCB. Clean pads with isopropyl 99% and a fiberglass pen. Verify pad integrity before soldering replacements - lifted pads on a multi-layer mains PSU PCB are very difficult to repair cleanly and almost always mean Tier 4 at D-Central's bench.

Install replacement caps with correct polarity. The band on the cap marks the NEGATIVE terminal. Backwards installation = explosion under power, often violent enough to damage adjacent components. Double-check every cap before the first power-on. Solder joints should be clean cones, not blobs - reflow with flux if needed. Trim leads short to avoid shorting against the chassis lid. Photograph the rebuild for your maintenance log and for D-Central's repair-queue dataset if you choose to share.

Replace the PSU's internal cooling fan if it shows any rattle, slow start, or wear during the rebuild. Typically 60 mm or 80 mm axial, 12 V, 2-pin or 4-pin. Match dimensions, voltage, CFM rating exactly. Dual-ball-bearing only - sleeve-bearing consumer fans die inside 90 days in the >80 °C interior environment. A failing fan cascades into bulk-cap failure within weeks. Fan parts cost $15. Cascaded cap-and-MOSFET failures cost $200+. Verify-flag: confirm exact fan PN against your specific PSU revision before ordering.

Bench-test the rebuilt PSU before reinstalling. Programmable DC load, 4 kW capable, run the PSU at full miner draw (~3,400 W, ~280 A on +12 V for KS5L) for at least 60 minutes continuous. Verify the rail holds 12.0 V +/- 0.2 V. If the rail sags at 200 A or trips at 250 A, the rebuild is not done - go back and check ESR, MOSFETs, and feedback loop. Verify-flag: home rebuilders rarely have a 4 kW programmable load. This is one of the strongest arguments for shipping a flaky PSU to D-Central instead of guessing whether the rebuild is bench-good at home.

Stop DIY and ship to D-Central when: rebuild trips at any load level after caps are healthy, primary-side short measured after parts replacement, transformer windings read open or shorted, lifted pad during rework, PCB has visible thermal damage / discoloration / charring, burnt-electronics smell (stop now, fire risk), or you are inside the IceRiver warranty window (file the RMA first, do not void warranty). Book a slot at https://d-central.tech/services/asic-repair/. Western English-language IceRiver repair authority. Canadian / US / international welcomed.

What D-Central does at the bench for KS-series cap rebuilds. Diagnostic against a known-good KS5L / KS5M reference rig with a logged AC source. Component-level rebuild including bulk cap replacement (Nichicon PW or Rubycon ZLH per current parts catalog), all secondary-side filter caps, internal fan, controller IC if needed, and full reflow on heat-fatigued joints. Programmable DC load test at full miner draw for 2 hours. Whole-miner burn-in at nameplate hashrate (~12 TH/s KS5L, ~15 TH/s KS5M) for 24 hours with rail-voltage logging. We document the rebuild, photograph each replaced cap, and ship with a printed test report. Verify-flag: nameplate hashrate values are directional and depend on firmware revision and ambient.

Ship the PSU separately or the whole miner - your call. PSU only: anti-static bagged, double-boxed with 5 cm foam on every side, printed note inside with AC voltage logs (if you collected them), observed symptoms, firmware version, and contact info. Whole miner: ship if you're unsure whether the fault is PSU-side, hashboard-side, or controller-side, and let the bench triage it. Canadian customers ship to our Quebec bench, expected turnaround 5-10 business days. US and international welcomed - contact the repair desk for international shipping paperwork. Optional: ask us to add the failed unit to D-Central's IceRiver bad-batch dataset before disposal, no charge for the documentation.