Whatsminer M21S – Low Hashrate Per-Board Diagnosis

Cold-boot at the wall for 60 seconds. Pull the cord, let the PSU bulk caps drain, reconnect. Some BTMiner builds latch a per-chain TH/s reading across soft reboots that doesn't reflect current state; a full cold boot forces the firmware to re-baseline. Re-check per-chain TH/s on the dashboard 15 minutes after boot. If the spread closes, you caught a transient; set up monitoring on per-chain delta > 5% and move on.

Verify intake ambient ≤ 30 °C with an IR thermometer at the front grille. A summer garage at 35 °C+ will trigger silent per-chip thermal derating across all three chains. If the throttling hits the chains unevenly (which it does — middle board typically runs hotter on M21S airflow geometry), you'll see a per-chain split that looks board-specific but is actually thermal. Fix airflow first. If the spread closes after ambient drops, the room was the problem.

Clean the intake grille and front heatsink face. Shop-vac with a soft-brush attachment. Six-year-old M21S chassis frequently arrive at the bench with intake heatsink fins packed solid with dust. Boundary-layer airflow at the chips on the dragging chain (often the most-blocked chain) drops first — and per-chain TH/s drops with it. A 60-second vacuum across the full intake face often closes a 15% per-chain spread on its own.

Pull the BTMiner log via WhatsminerTool and check for sibling codes. If the log shows 350-352 (hashboard high temp), 233-238 (PSU output protection), 600/610 (ambient too high), or 550-552 (bad chips) on the dragging chain, address those first — they're the upstream cause and the per-chain split is the downstream symptom. A standalone per-board drag with no sibling code in the log is the case where Tier 2+ work is justified.

Compare per-chain TH/s to nameplate-divided-by-three baseline. M21S 56T = ~18.6 TH/s per chain healthy. M21S 54T = ~18 TH/s. M21S 52T = ~17.3 TH/s. M21S 50T = ~16.7 TH/s. Confirm which bin you actually own — secondhand listings frequently mis-label bins, and 'low hashrate' sometimes turns out to be 'this is actually a 50T chassis sold to me as a 56T.'

Cold-boot and reseat the ribbon cable on the dragging chain at both ends. Power off at the wall first. Disconnect the ribbon at the board side and the control-board side. Inspect both connectors for oxidation, bent pins, dust packing, or visible mechanical damage. Wipe contacts with 99% IPA on a lint-free wipe; let dry; reconnect firmly until you hear the connector click. Power up, observe per-chain TH/s for 15 minutes. Roughly 10-15% of 'per-board' complaints clear here — the cable was marginal, not the board.

Reseat the per-chain power harness. Same discipline: power off at the wall, disconnect the heavy-gauge power leads from the dragging board, inspect contacts for oxidation or charring, reconnect tight. M21S board-power connectors that have been cycled multiple times (secondhand chassis frequently have) develop intermittent contact resistance that drops board voltage under load and drags chain TH/s.

Slot-swap the dragging board into a known-good slot. Label slots 0/1/2 with masking tape so you don't lose track. Rotate the suspect board into a working slot, rotate a healthy board into the slot the suspect came from. Cold-boot. Watch per-chain TH/s for 15 minutes. This is the diagnostic that proves board-vs-slot. If the drag follows the board into the new slot → board-level fault, escalate to Tier 3. If the drag stays in the original slot regardless of which board now sits there → adapter or slot-side issue.

Measure the 12V rail at the dragging board's power input under load. Multimeter on DC, probe at the M21S board-power connection while the chassis is hashing at full power. Expect 11.8-12.6V sustained. Below 11.5V at this single board while the other two boards measure clean = local distribution issue (oxidised lug, marginal PSU output, broken control-board sequencing on that channel). Above 12.6V is also a problem (regulator runaway). The rail measurement at the board, not at the PSU, is the one that matters for per-board diagnostics.

Reset the firmware profile to stock. WhatsminerTool → Remote Ctrl → reset config / Power Fast Boot. Stock profile, observe 30 minutes. If per-chain spread closes, profile was the cause — the board was being asked to do something the chips couldn't sustain on aged silicon. Either accept the lower nameplate or rebuild the profile slowly with stability windows between steps.

Pull chipcount per chain via the BTMiner API on TCP 4028. Command: echo '{"cmd":"summary"}' | nc <miner-ip> 4028 and echo '{"cmd":"devs"}' | nc <miner-ip> 4028. The devs response gives per-chain chip status. Healthy M21S = 105 chips per chain. A dragging chain reporting 99 / 102 / 103 chips confirms dead-chip accumulation explicitly. Below ~95 chips on a chain, the board is at end-of-economic-life unless you want to invest in chip-level rework — see Tier 3 / Tier 4.

Thermal-image the suspect board under load. FLIR or Seek camera, side-on shot through an open chassis (fans temporarily removed, miner hashing at reduced power). Healthy BM1397 chip array = uniform 60-75 °C across the board. Look for cold spots (chips at 30-40 °C while neighbours are hot) — those are dead chips. Look for cold contiguous regions of 4-8 chips — those are dead voltage domains. Map the cold region against the M21S board layout.

Apply fresh thermal paste to the chips on the suspect board. Pop the heatsink, IPA the old paste off both the chip tops and the heatsink underside, apply a uniform thin layer of Arctic MX-6 or Thermal Grizzly Kryonaut, reassemble with even torque on the heatsink fasteners. M21S boards that have run for 3+ years on factory paste frequently recover full per-chain TH/s after a paste refresh — sometimes the entire 15% shortfall is paste, not chips.

Inspect voltage-domain capacitors and MLCCs under magnification. USB microscope or 30x loupe across the buck-regulator and PMIC regions of the suspect board. Bulging electrolytics, cracked MLCCs near a buck-boost inductor, dry-solder joints on the regulator IC pins — all of these cause domain undervoltage, drop chips out of the hashing cluster, drag chain TH/s. This is a soldering-iron + hot-air repair, not a reflow job.

Reflow individual dead chips identified by thermal imaging. If thermal imaging shows one or two specific chips reading cold while neighbours are hot, those chips are likely victims of solder-joint fatigue from years of thermal cycling. Hot-air rework station, preheat the bottom of the board to ~150 °C, top-side hot air at 310-330 °C for ~30 seconds, let cool naturally. BM1397 packages tolerate a single reflow cycle reasonably well. A second reflow on the same chip rarely helps — at that point the chip is replacement-only.

Chip-replacement attempt with matched-bin BM1397 salvage. If reflow doesn't recover a specific dead chip, the only DIY-adjacent option is replacement with a matched-bin BM1397 from a parts donor (typically a junked M20S or M21S board). This is bench-grade work — hot-air station, fine flux, solder paste stencil for the BGA, post-rework cleanup. If you don't have rework experience on 0.4mm BGA packages, skip this and ship the board to a bench. The cost of one botched chip replacement can write off a board that was otherwise recoverable.

When to stop DIY. Any of: (a) Tier 3 thermal paste refresh didn't recover the chain, (b) cold voltage-domain region visible on thermal imaging, (c) chipcount on the dragging chain has dropped below ~95, (d) you don't have hot-air rework, thermal imaging, or matched-bin BM1397 salvage stock. The economics matrix matters here: a six-year-old M21S 56T at 2026 secondhand pricing is $150-$400 USD. A bench-grade chip-replacement repair runs $95-$275 CAD per board. If the board is borderline, sometimes the right answer is to sell the chassis as-parts and reinvest the recovered cash in a current-generation miner. Book a D-Central ASIC Repair slot for the diagnosis call before committing.

What D-Central does at the bench on M21S. Per-chain test on a programmable load with known-good control board and known-good PSU eliminates slot/adapter/PSU as the variable. Per-chip voltage and frequency sweep under calibrated reference identifies exactly which chips are dead, marginal, or out-of-spec. Chip replacement from matched-bin BM1397 salvage stock (we keep parts donors specifically for the M20/M21 generation). Voltage-domain rework when the cold-region thermal signature points there. Fresh paste, reseal, reassembly. BTMiner reflash against the correct hardware-revision build. Then 24-hour burn-in at nameplate in a 25 °C controlled environment to confirm the repair holds. Turnaround 5-10 business days, Canada-wide; US and international welcome.