Antminer – Undervolting Too Aggressive

Revert to stock voltage. Disable any custom UV profile, disable autotune, reboot, and let the miner run 15 minutes. If HW% drops under 1.5%, your undervolt was the cause — rebuild it slowly in Tier 2. This is the single fastest diagnostic in the entire sequence and it costs nothing. Ninety percent of 'mystery HW%' tickets end here.

Hard power-cycle at the breaker for 30 seconds. Not a soft reboot — a real cold start. Clears any wedged driver state from a recent firmware or tuning change and gives the NVRAM a clean read on the next boot. A one-shot post-tuning glitch can end here without deeper investigation.

Verify ambient temperature ≤ 35 °C at the intake grille with an IR thermometer — not room-middle, not the hallway, at the intake. Ambient heat shrinks the stable-voltage window: a chip that was fine at -50 mV in January will throw HW% at -50 mV in August. Fix ambient first, undervolt second.

Clean the intake filter and verify nothing is blocking the front 15 cm of the miner. Dust on the filter = higher inlet temp = narrower stable window. Shop-vac the filter, wipe the grille, and confirm no furniture, curtains, or dust buildup around the intake.

Check Bitmain firmware version at service.bitmain.com/support/download against your control-board hardware revision sticker. If you're on a build known for autotune regressions on your revision, roll one version back or forward as a Tier-1 diagnostic. Bitmain acknowledges autotune crashes on specific chip bins (https://support.bitmain.com/hc/en-us/articles/18237912339097).

Measure PSU output at the board connector under full load. Multimeter on DC, probes on the PSU-to-hashboard connector while the miner is hashing at nameplate. Expect ≥ 13.8 V sustained on a standard S19, ≥ 13.6 V on S17, ≥ 12.8 V on S9. Below that = PSU tired or circuit undersized. Swap PSU with a known-good unit before suspecting firmware; an undervolt that fails on a sagging PSU will appear fixed once the rail is clean.

Measure line voltage at the panel under load. 240 V split-phase should read 235-245 V; 208 V commercial should read 202-212 V. If you're on 110 V and it's an S19 or newer, that's your answer before any undervolting discussion begins. Low line voltage means high PSU current draw, which means real-world rail voltage is lower than your firmware setpoint.

Re-seat every hashboard data and power cable. Power off at the breaker. Disconnect each connector, inspect for blackening, bent pins, or corrosion, reconnect firmly — listen for the click. A cable that's almost seated produces intermittent voltage on one domain that looks exactly like an aggressive undervolt but is actually a connector problem. The click matters.

Rebuild the undervolt from stock in -20 mV steps — half the community default step size. After each step, watch HW% for 15 minutes. Stop one step before HW% crosses 1.5%. That's this miner's stable floor. Write it down: frequency, voltage, ambient temp at the time, line voltage. You will re-tune every 90 days as chips age and seasons shift. The floor is not static.

Verify the PSU cord and PDU are correct for your model. S21 uses a new P13 20 A cord (https://support.bitmain.com/hc/en-us/articles/27522322154521); an undersized S19 cord on an S21 or a damaged cord on an S19 produces sag that mimics aggressive-undervolt symptoms. Use the cord the miner shipped with.

Flash DCENT_OS — D-Central's own open-source Antminer firmware. Per-chip HW%, per-domain voltage adjust, closed-loop autotune (steps up and observes per chip rather than trusting an open-loop V-F table), stratum v2, open-source, no licensing fees, Mining-Hacker-maintained. https://d-central.tech/dcent-os/ (landing) and https://github.com/DCentralTech/DCENT_OS (source). Let the miner stabilize 20 minutes after flash, then inspect per-chip HW%. This is the single most valuable diagnostic upgrade on any Antminer you mine with seriously.

Alternative firmwares if DCENT_OS isn't yet available for your hardware revision: Braiins OS+ (https://braiins.com/os/plus), LuxOS (https://luxor.tech/luxos), or Vnish — all expose per-chip HW%. Braiins OS+ has a useful reversible trick on some models: it runs from a microSD card, so popping the card reverts to stock (https://altairtech.io/product/braiins-antminer-microsdc/) — zero-risk diagnostic. Note custom firmware voids Bitmain warranty, a non-issue on out-of-warranty units.

Use per-domain voltage autotune on DCENT_OS or Braiins OS+. Each of the 38 domains on an S19 now settles at its own stable floor instead of a single global floor set by the weakest domain. Braiins documents 10-15% efficiency gains over stock at equivalent hashrate and 38-45% power savings at underclocked 80 TH/s vs stock 110 TH/s (https://braiins.com/blog/smarter-clock-speed-management-for-bitcoin-miners-maximize-roi-minimize-waste). This is the correct way to do aggressive undervolting — per-domain, closed-loop, with per-chip HW% telemetry proving it works.

Disable specific failing chip positions on DCENT_OS / Braiins OS+ / LuxOS. Lose that chip's ~0.9 TH/s (on an S19), keep the rest of the board running at your target undervolt. For a board with one chip that sets the floor 40 mV higher than the other 75, disabling is usually the right call — you'd lose more hashrate running the whole board at the higher voltage than you'd lose disabling one chip. Disable until 2-3 chips on the same board are gone; then replace the board.

Refresh thermal paste and pads. Dried paste or crumbled pads narrow the stable-voltage window measurably — a 2-3 °C junction-temp increase moves the silicon's V-F curve enough to kill an undervolt setpoint that was fine three seasons ago. Arctic MX-6 or Thermal Grizzly Kryonaut, uniform thin layer. Pay special attention to the PCH and voltage-domain ICs where dried pads are most common.

Replace the CR2032 RTC battery on the control board for any S19 more than 2 years old. A dead RTC battery can cause TLS cert expiry checks to fail during pool auth; some firmwares cascade that into a service crash that looks like undervolt instability. $2 fix, 5-minute job — no reason to skip it if you're in the chassis anyway.

Stop DIY and book a D-Central repair when per-chip HW% isolates a chip that's 30+ mV higher on its stability floor than the rest of the board (and you don't want to disable it), OR the PMIC is suspected, OR you pushed an undervolt hard enough to kernel-panic into a boot loop and SD-card recovery isn't landing. https://d-central.tech/services/asic-repair/ · turnaround 5-10 business days · Canada-wide shipping · US and international welcomed.

Ship hashboards in anti-static bags, double-boxed with ≥ 5 cm foam on every side. Include a one-page note: observed symptoms, firmware version and build string, the exact voltage and frequency setpoint at failure, ambient temp at the time, PSU model, and your contact info. Diagnostic time at the bench is repair cost — a tight ticket saves both sides time and money.