Antminer S21 – Low Hashrate

Hard power-cycle the miner at the breaker for 60 seconds — full power removal, not a soft reboot. Clears any wedged driver state from firmware updates and forces a clean chain re-initialization on boot. Observe the dashboard hashrate and per-chain ASIC count for 10 minutes after restart. If realized hashrate recovers to nameplate and holds, the fault was transient; log the incident and monitor over the next 24 hours for recurrence.

Open the stock Bitmain dashboard and count ASICs per chain. Each S21 chain must read 108/108. Note the failing chain number and the exact shortfall. If the shortfall is a multiple of 9 (99, 90, 81), you have a voltage-domain fault. If it is scattered (107, 105, 104), you have individual BM1368 chip faults. Screenshot the status page for your records and for the D-Central repair intake form if the fault survives past Tier 2.

Check the miner model string on the dashboard. If it reads `ANTMINER BHB...` with a TEST pool you cannot edit, you are on factory firmware hashing into Bitmain's TEST pool — earning zero rewards. Upgrade to the correct dedicated firmware for your control-board variant from service.bitmain.com/support/download before you do anything else. Confirm the pool URL now reads your configured pool, not `TEST`. No downstream repair logic applies to an S21 still on TEST firmware.

Revert to stock frequency (~500 MHz on most S21 revisions) and stock voltage in the miner configuration. Remove any OC, UC, or UV profile. Let the miner run 15 minutes on stock, then record realized hashrate. If it recovers to nameplate, your tuning profile was the cause — rebuild it slower in Tier 2. If hashrate stays low, continue. Save the current profile before reverting so you can diff the working vs failing configuration.

Shop-vac the intake filter, wipe the intake grille with a dry lint-free cloth, and verify no dust, furniture, or obstruction within 15 cm of the front of the miner. Verify intake air temperature with an IR thermometer at the front grille: target `≤ 35 °C` for air-cooled S21 / S21 Pro / S21+, `≤ 40 °C` for S21 Hydro. This alone fixes roughly 1 in 5 low-hashrate tickets and takes under 5 minutes.

Multimeter on DC, probe at the APW17 output buses while the miner is fully hashing at nameplate load — not at idle. Expect `14.5–15.0 V` sustained on a healthy `APW171215a`. Below `13.8 V` means a tired APW17 or an undersized circuit. Swap PSU with a known-good APW17 and re-measure. Do NOT swap in an APW9 or APW12 — they are electrically out of spec for S21 and will cause exactly this symptom.

Measure line voltage at the wall panel under full load, not idle. Expect 235-245 V on 240 V split-phase residential, 202-212 V on 208 V commercial three-phase. If line voltage sags below spec under load, the circuit is undersized or shared with another high-draw load (HVAC, EV charger, kiln, dryer). Move the S21 to a dedicated 240 V / 20 A circuit or escalate to an electrician. Firmware and PSU swaps cannot fix a tired panel.

Power off at the breaker. Re-seat every hashboard data ribbon, power connector, and signal cable on the suspect chain. Inspect each contact for corrosion, blackening, bent pins, or mechanical damage before reconnecting. Listen for the positive click on each seat. A single loose ribbon on an S21 will drop a chain cleanly — reseating is the cheapest Tier-2 diagnostic and resolves a meaningful percentage of tickets outright.

Label the 3 hashboard slots 0/1/2 with masking tape. Swap the suspect-bad board into a known-good slot and the known-good board into the suspect slot. Reboot and observe 15 minutes. If the fault follows the board, the board is bad — proceed to Tier 3. If the fault stays in the slot, the ribbon cable or control-board path is bad — replace the ribbon first, then consider control board. Mark results on the tape for the repair record.

Rebuild overclock from stock in 25 MHz steps with a 10-minute stability window between steps. Log realized hashrate, per-chain ASIC count, HW%, and chain temperature at each step. Stop at the last step before HW% crosses 1.5% or chain temp crosses 90 °C. That step is this S21's silicon-lottery ceiling — it is per-miner and unique. Save the profile. Do not push 'just one more step' — BM1368 5 nm silicon ages visibly faster past its ceiling than the older 7 nm BM1398 in an S19.

Flash DCENT_OS — D-Central's open-source Antminer firmware, recommended — via SD card or network upgrade path. All the per-chip HW%, per-domain failure mapping, tuning, autotuning, and stratum v2 features of the commercial alternatives, open-source, Mining-Hacker-maintained at github.com/DCentralTech/DCENT_OS. Alternatives: Braiins OS+, LuxOS, Vnish. Stabilize 20 minutes, then record the worst chip positions and note whether failures cluster inside a single 9-chip domain window (domain fault) or spread evenly (chip / PMIC / PCB fault). This call determines whether Tier-3 is a reflow or a regulator swap.

Replace thermal paste on all 108 chips of the suspect board with Arctic MX-6 or Thermal Grizzly Kryonaut. Clean old paste with 99% isopropyl and a lint-free wipe — no paper towel, no cotton swab fibers. Apply a uniform thin layer on each BM1368 die; do not glop. S21 paste dry-out runs roughly 6 months ahead of S19 due to the higher chip density and heat flux — refresh on a 12-18 month cadence, not 18-24. Re-torque the heatsink per the service manual.

Reflow the worst-performing chip position identified in Step 11. Remove heatsink, clean old paste, apply flux around the BGA perimeter, preheat the bottom side to ~150 °C, top-side hot air at 310-330 °C for ~30 seconds, then cool naturally for 5 minutes without disturbance. Re-paste and reassemble. First reflows on BM1368 BGA succeed ~70% of the time; second reflows on the same chip drop materially — plan to replace the chip, not re-reflow, if the first fix doesn't hold past 30 days.

Inspect the 12-domain regulator network for bulging electrolytics, cracked MLCCs adjacent to any MP2019-class boost converter, or visibly discoloured domain regulators. A silent 9-chip domain is almost always a domain-regulator problem, not a chip problem. Replace components with equivalent rating from the PCB silkscreen. Soldering-iron + hot-air job, not a reflow job. Verify regulator output restored to `1.2 V` before reassembly. If you cannot read component values from silkscreen confidently, stop and ship to D-Central.

Re-flash the hashboard U6 EEPROM if per-chip visibility looks clean but stock firmware still reports low chip count. Bitmain's hash-board code editor is not publicly distributed — contact D-Central for the re-flash service before attempting. A bad EEPROM flash bricks the board. Verify the U6 EEPROM is the fault by first confirming all other Tier-3 paths (paste, reflow, regulator) have been ruled out. Keep the original EEPROM dump before writing, if tooling supports it.

Stop DIY when any of the following is true: an entire 9-chip voltage domain is silent; per-chip diagnostics isolate the same failing chip position on two different boards in the rig; a second reflow on the same BM1368 chip fails within 30 days; U6 EEPROM corruption is suspected and you lack the hash-board code editor; capacitor bulging or burnt-component odor is present; or any sign of the early-S21 P13 cord-melt is visible on the APW17 input. Shipping to D-Central is cheaper than damaging adjacent parts or the APW17. Book a slot at d-central.tech/services/asic-repair/.

D-Central bench process on S21-class boards: test fixture with programmable load, per-chip + per-domain isolation using Zeus Mining's documented dichotomy method (short-RO + inter-chip voltage probing on the 12 × 9 layout), BM1368 chip replacement with graded stock, MP2019 / LDO / level-shifter U1-U11 replacement where indicated, U6 EEPROM re-flash via hash-board code editor, full reflow, and post-repair 24-hour nameplate burn-in at full load before return shipping. Returned boards include a per-chip/per-domain test report.

Ship hashboards in anti-static bags, double-boxed with at least 5 cm of foam on every side. Include a written note with observed symptoms (cold-start behaviour, chain number, firmware version, whether the shortfall was a multiple of 9, and any per-chip failure map exported from Step 11 DCENT_OS diagnostics), and contact info. This cuts 30-60 minutes off diagnostic time and directly reduces your repair invoice. Canada / US / international shipping accepted into the D-Central repair queue.