12V Rail Sag — Hashboard Brownout Under Load

Confirm one PSU output port → one PCIe cable → one hashboard, no daisy chains. Power off and trace every PCIe cable from PSU to hashboard. If a single Y-split cable feeds two hashboards, or two hashboards share a daisy connector off one PSU port, that is the failure. Replace with one dedicated PSU output port → one cable → one hashboard. The 150-W-per-PCIe rule is a connector-physics limit (Mini-Fit Jr. pins rated `~6 A` per circuit), not a guideline. Most S19/S21-class boards have two 6-pin inputs by design — feed both.

Replace cheap or aged miner cables with 16 AWG (minimum) miner-rated PCIe sets. Many 'deal' kits ship 18 AWG marketed-as-16 AWG; verify by stripping `1 cm` and measuring conductor diameter — 16 AWG is `~1.29 mm`. For S19 / S21-class loads at over 1000 W per board, 14 AWG is the right answer, not the optional answer. Keep cables as short as workable — every metre adds about `5 mΩ` of resistance per conductor at 16 AWG, which steals headroom from the rail at the hashboard end.

Confirm the PSU is on a dedicated wall outlet on a dedicated breaker, not stacked on a power strip or surge bar. Miner PSUs draw too much current to share a 15 A power strip — they will under-volt, the strip will heat, the breaker will trip, or all three. If you're sharing, separate them. If the run is long, use 12 AWG line-side conductor or heavier — the wall-to-outlet resistance counts against your headroom too.

DMM-check 12V at the miner end under load: multimeter on DC, probe right at the 6-pin going into the hashboard with the miner hashing flat-out. Healthy: `12.0–12.6 V`. Marginal: `11.7–12.0 V`. Sagging: below `11.7 V` at the connector under load means brownout territory regardless of what the PSU display reads. If you have min/max capture on the DMM, use it for at least 5 minutes — average DC reading lies, the sag is in the trough.

Reboot once and confirm the symptom returns under load. Some 'rail sag' reports are actually one-time PMIC latches that clear with a reboot. If the symptom doesn't come back, monitor for 48 hours before declaring it fixed — but don't skip the cable / PSU verification regardless. A latch that cleared once will latch again.

Scope the rail at the hashboard connector for 60 seconds under full load. AC-coupled, 100 mV/div, 1 ms/div timebase. Capture peak-to-peak ripple and any deep dips. Healthy ASIC PSU: `< 120 mV pp`. Degraded: `> 300 mV pp`. Catastrophic: spikes dipping more than `0.6 V` below DC mean. Spikes that dip below the PMIC UVLO threshold confirm the brownout chain. If you don't own a scope, an FNIRSI 1014D is `~$200 CAD` and a used Rigol DS1054Z is `~$300 CAD` — cheapest credible options. A scope is the single tool that catches a 200-microsecond rail dip a DMM averages away.

Log line voltage at the wall outlet for 24 hours with a Kill A Watt or equivalent plug-in logger. Capture min / max. Healthy 240 V circuit stays above `225 V` minimum; healthy 120 V stays above `108 V`. If line dips correlate with miner crash log timestamps, the circuit is the constraint, not the miner. Plan an electrical upgrade — dedicated 30 A or 50 A breaker, heavier conductor, PDU on a beefier circuit.

Move to a 240 V dedicated circuit if currently on 120 V. A single S19 on 120 V draws `27 A` at full load — over a 30 A circuit derate. The same miner on 240 V draws `13.5 A`. Half the current, half the heating in every connector and conductor in the path, double the headroom. This is not a luxury for serious mining; it's table stakes. Cost-wise, an electrician dedicated 240 V run is `$200–600 CAD` in most North American jurisdictions.

Replace any cable that runs warmer than `50 °C` under load. IR thermometer on the cable jacket while the miner hashes. Anything past warm-to-the-touch is dropping too much voltage as heat — that voltage is being stolen from the hashboard PMIC. Hot cable equals sagging cable, every time. Replacement cables should be 16 AWG minimum for 800 W loads, 14 AWG for over 1000 W per hashboard, with verified copper conductor and not copper-clad aluminum (CCA — rejected on conductivity, ages badly, dangerous at miner currents).

Tighten every screw terminal on a server-PSU breakout board with a torque driver to spec (most boards spec `1.5–2.0 Nm` on M3/M4 output studs). Loose terminals are an enormous source of contact resistance — micro-arcing under load progressively destroys the contact surface, and rail sag worsens weekly. Annual torque-check is the cheapest preventive maintenance you'll ever do on a breakout-based build. Use a calibrated torque driver, not a guess-by-feel Phillips screwdriver.

Replace dried output bulk capacitors in the PSU. Cover off, bulk cap discharge with a `10 kΩ 5 W` resistor across the bulk cap terminals for 30 seconds, verify `< 10 V` on a DMM before touching anything. ESR meter (DER EE DE-5000 or equivalent) on each output-side electrolytic. Healthy: `< 30 mΩ`. Marginal: `30–100 mΩ`. Failed: `> 100 mΩ`. Replace any failing cap with same-or-better voltage rating, same-or-larger capacitance, low-ESR series (Panasonic FR / FM / FC, Nichicon PW, United Chemi-Con KZE / KZH). On Whatsminer P21 / P21B this is the fix that resolves a startling proportion of 'PSU is dying' tickets — the PSU is fine, the caps are dry. On APW9 / APW12 expect 4–8 caps to be near end-of-life by year 4 of continuous duty.

Replace input filter capacitors on the hashboard if rail at the connector is clean but the PMIC is still tripping. SMD electrolytics or polymer caps near the 6-pin connector. ESR-test in-circuit (rough but workable), or desolder for an off-board reading. Polymer replacements (Panasonic POSCAP, Sanyo OS-CON) are dimensionally compatible and dramatically improve transient response. Recommended over electrolytic if your hot-air station can place them cleanly. Reflow at `350 °C` top-side, soak the board at `120 °C` bottom-side preheat to avoid pad lift.

Add a bulk capacitor at the hashboard end of the cable for severely marginal builds. A `2200 μF / 16 V` low-ESR electrolytic across 12 V / GND at the hashboard connector terminal, on its own short leads, gives the rail a local reservoir for spike currents. This is a workshop band-aid, not a long-term fix — it masks the cable problem rather than solving it. Use it to confirm the diagnosis (symptom disappears = sag was real, cable is undersized), then replace the cable. Don't ship a production miner with a tacked-on cap.

Re-pin or re-crimp marginal PCIe connectors. A backed-out female pin in a 6-pin Mini-Fit Jr. connector is invisible until you get a pin extractor in there. Molex 11-03-0044 extractor for Mini-Fit Jr.; gentle work, don't force. Reseat or replace female terminals with fresh ones from a Molex crimp kit. A `$20 CAD` fresh terminal is cheaper than a `$400 CAD` hashboard. Inspect the connector body for cracking, discolored plastic, or carbon tracking — any of those = replace the entire connector, not just the pin.

Server-PSU breakout builds: upgrade to a multi-stud current-sharing breakout (Parallel Miner X11, Bitronics, or equivalent) fed from an HP DPS-2400 or Delta DPS-2000. Cheap eBay breakouts crowd 200 A onto a single stud and melt under any meaningful load. Multi-stud designs distribute load across 4–6 output bars, each rated for tens of amps. The expensive breakout pays for itself the first time it doesn't melt under a Bitaxe Hex farm or a small Antminer build. Wire each output stud to a dedicated load with the same gauge / length / connector quality discipline as a manufacturer build.

Stop DIY when: multiple PSU output capacitors fail simultaneously (full re-cap is faster as a shop job), the breakout board PCB shows scorching or lifted traces, on-board hashboard input filters need rework that exceeds your hot-air competence, or you've replaced cables, swapped PSUs, and re-capped and the symptom persists. At that point you need a programmable bench load with calibrated transients to nail it down, which is bench gear most home miners don't have. Book a slot at https://d-central.tech/services/asic-repair/ — flat-rate diagnostic, parts at cost.

Ship clean: pack the PSU, suspect cables, and hashboard separately, anti-static bag the boards, double-box with `≥ 5 cm` of foam on every side. Include a note with observed symptoms, wall-side line voltage during failure, scope captures if you took any, firmware version, miner model and revision, and contact info. Diagnostic time saved at the bench equals bench cost saved equals your repair bill saved. The shop will reach you with a flat-rate quote before any parts work begins.