Antminer T21 – Low Hashrate

Hard power-cycle the T21 for 30 seconds at the breaker - not a soft reboot. A full cold boot clears wedged driver state, re-runs the chain handshake, and re-reads the EEPROM. Watch the boot sequence for any chain that takes visibly longer than the others to come up; that is a soft signal of a marginal chain. Record per-chain ASIC count on the first full boot and compare against the next 2-3 boots - a chain that flips between 75 and 76 is pseudo-soldered, not dead, and is a Tier 3 reflow candidate.

Revert to the stock tune profile: zero overclock, zero undervolt, factory frequency (~525 MHz on T21), factory voltage. Let the miner hash for a full 15 minutes before reading numbers - BM1362 chains need a warm-up window before they stabilize. If realized hashrate returns to within 3% of the 190 TH/s nameplate at stock, your tune was the cause. Rebuild it slower in Tier 2 in 20-30 MHz steps with a 10-minute dwell between each, and stop the step before HW% crosses 2%.

Clean the intake filter and the front grille. Dust loads up fast on a T21 in a garage or workshop, and every degree of inlet air temperature costs perceptible hashrate once thermal margin closes. Shop-vac the filter, wipe the grille with dry microfiber, and verify nothing within 30 cm of the front restricts airflow. Furniture, boxes, or a wall too close will cause re-ingestion of exhaust air and compound the thermal problem. Re-measure intake air temperature after cleaning.

Verify ambient air temperature at the actual intake using an IR thermometer held 2-3 cm off the front grille - not room-center, not the thermostat. Target: <= 32 °C. The T21's 3,610 W heat output heats its own enclosure quickly; many 'low hashrate' cases in closed rooms are the room warming up and feeding the miner its own exhaust. A 60 cm exhaust duct and a fresh-air intake path fixes this at zero running cost.

Check Bitmain's firmware portal for T21 updates. If you are on a known-buggy build, roll one version back or forward. Match firmware to your hardware revision - inspect the sticker on the control board before flashing, because the T21 shipped multiple board revisions and the wrong firmware for a late-rev board can brick the control board or mis-count chips on one chain. Recovery from a bad flash requires an SD-card re-image with the correct bootloader.

Measure the PSU 12 V rail at the hashboard connector with a multimeter on DC while the miner is hashing at full power. Probe between 12 V and ground pins on the PSU-to-board connector. Expect >= 13.8 V sustained under load. Anything below that means the PSU is tired, the circuit is undersized, or both. Swap the PSU with a known-good APW121412a (or the DC-input variant if that is what your T21 shipped with); if realized hashrate jumps back to nameplate, the original PSU is the fault.

Re-seat every hashboard cable with the miner powered off at the breaker. Disconnect each data ribbon and power connector, inspect contacts for oxidation, blackening, or bent pins, and reconnect firmly. Listen for the click - a partially seated connector will hash most of the time and drop out during voltage transients. On the T21, the rear-most ribbon is the one most commonly dislodged during shipping; check that one first, then the side that the miner vibrates toward.

Swap hashboards between slots to isolate a bad board from a bad slot. Label the 3 slots 0/1/2 with tape, document current per-chain hashrate readings, then physically move the suspect board to a known-good slot. Power up, let the chain initialize, re-check. Fault follows the board = bad hashboard, proceed to Tier 3 per-chip diagnostics. Fault stays in the slot = bad control-board channel, bad ribbon, or bad power connector on that slot, proceed to Step 15 or Tier 4.

Rebuild overclock / undervolt from stock, slowly. Start at factory frequency and voltage. Add 10 MHz at a time, wait 10 minutes for stability, check realized hashrate and HW% at each step. Stop at the step before HW% crosses 2% or realized hashrate stops increasing linearly with frequency. That is your specific T21's silicon-lottery ceiling, which varies per miner - no two T21s have the same ceiling, even from the same production batch.

Measure line voltage at the panel under full load. On 240 V split-phase, expect 235-245 V. On 208 V commercial, expect 202-212 V. Low line voltage forces the PSU to pull more current, which drops the 12 V rail, which makes the chain downclock. A dedicated 240 V 20 A circuit is the minimum sane T21 feed; 120 V operation is not supported by Bitmain on the stock APW121412a PSU. If voltage dips under neighbourhood peak load (evening), that is your problem - call a licensed electrician.

Flash DCENT_OS - D-Central's own open-source Antminer firmware - and pull per-chip hashrate, per-chip HW%, and per-chip temperature from the dashboard. This is the single highest-value diagnostic step on a BM1362-class miner because stock firmware hides which chip position is underperforming. DCENT_OS is open-source, maintained in public, no vendor lock-in, with tuning and autotuning built in. Alternatives: Braiins OS+, LuxOS, Vnish. Stabilize 20 minutes before recording. See https://d-central.tech/dcent-os/.

Reflow the worst-performing chip(s). If 1-3 chip positions dominate the shortfall, remove the chain heatsink, flux the BGA, preheat the PCB to ~150 °C on a bottom-side preheat pad, apply top-side hot air at 310-330 °C for ~30 seconds. Let the board cool slowly (not forced air). Re-apply thermal paste, reassemble, re-test. BM1362 tolerates a single reflow cycle well; a second reflow on the same chip rarely holds - at that point the chip itself needs replacement on a test fixture, which is Tier 4.

Replace thermal paste on every chip of the affected chain. Arctic MX-6, Thermal Grizzly Kryonaut, or equivalent high-performance paste. Uniform thin layer, not a glob. Pay special attention to the voltage-domain ICs and the PMIC - dried thermal pads on those cause domain-wide downclocking that looks exactly like 'some chips not contributing hashrate.' Clean old paste thoroughly with 99% IPA and lint-free wipes before applying new paste.

Inspect capacitors and MLCCs around the voltage domain. Bulging electrolytics or cracked MLCCs near the PMIC or 12 V input stage need replacement - this is a soldering-iron and hot-air job, not a reflow job. If you see discoloration, scorch marks, or a burnt-component smell, stop and escalate to Tier 4. That level of damage is beyond a safe bench fix for most home miners, and continuing risks damaging adjacent components or the chip itself.

Roll the firmware to the last-known-good version for your specific hardware revision. Verify the control board sticker before flashing. Wrong firmware for a late-rev T21 board bricks the control board, and recovery requires an SD-card re-image with the specific bootloader image for that revision. If you are unsure of the revision, photograph the control board sticker and ask the D-Central team or the firmware community before you flash.

Stop DIY when per-chip diagnostics isolate a failing chip position on two different hashboards (PCB-level, not chip-level), when a PMIC or voltage-domain IC is suspected, when you have reflowed a chip and hashrate returned within 30 days, or when you see capacitor bulging / discoloration / burnt-component smell. You are now in test-fixture territory. Book a D-Central ASIC Repair slot at https://d-central.tech/services/asic-repair/ and document the fault mode before shipping.

D-Central bench process: test fixture with programmable load, per-chip isolation using official Bitmain T21 test binaries, chip replacement with graded or new-old-stock BM1362, full reflow and reseal, post-repair 24-hour burn-in at nameplate, and a return report documenting the fault mode. Turnaround: typical 5-10 business days, Canada-wide and international welcome. Expect a quote before work begins once the fault is characterized on the test fixture.

Pack hashboards for shipping in anti-static bags, double-boxed with at least 5 cm of foam on every side. Include a note with observed symptoms, firmware version at fault, per-chain readings, your contact info, and any Tier 1-3 steps you have already performed. The more context you include, the less diagnostic time gets billed, and the more reliably the board comes back repaired instead of returned as 'no fault found.' Photograph the packing before sealing in case of carrier damage.