MicroBT Whatsminer M31S

The MicroBT Whatsminer M31S is a 2020-generation SHA-256 ASIC that mines Bitcoin at roughly 76 TH/s while drawing about 3,220 W — an efficiency near 42.4 J/TH. It is a three-board, Samsung 8nm-class air-cooled workhorse, now a legacy-tier machine best run for heat recovery, hobby mining, or cheap secondary hashrate.

Whatsminer M31S at a glance

Chip and hashboard architecture

The M31S belongs to MicroBT’s M3x family — the 2020 generation built on Samsung 8nm-class SHA-256 silicon. Unlike Bitmain, which fabricates its BM-series chips at TSMC, MicroBT designs its hashing ASICs around Samsung foundry processes. That foundry choice is one of the defining differences between a Whatsminer and an Antminer of the same era, and it carries all the way down to how the boards are wired, powered, and repaired.

Inside the chassis sit three identical hashboards feeding a single controller. Each board carries a long string of Samsung 8nm-class ASICs; the closely related M30S populates 148 chips per board for 444 chips total, and the M31S follows the same three-board MicroBT layout. Power is delivered to the chips through voltage domains — groups of chips regulated together — rather than one rail per chip. This per-domain control is a hard architectural fact worth understanding before any tuning or repair: you cannot dial an individual chip’s voltage, only the domain it sits in, which is why a single weak chip can drag a whole domain.

The control board: ARM, not FPGA

This is where Whatsminer diverges most sharply from Antminer. An Antminer S9 or S19 pairs an ARM core with a Xilinx Zynq FPGA that runs a hardware “midstate” engine to feed work to the chips. The M31S has no FPGA. It uses a pure Allwinner ARM system-on-chip — the M30/M31 generation rode the quad-core 64-bit H6 (ARM Cortex-A53) on MicroBT’s CB4-class control board, with higher-grade units moving to the H6OS / CV200-OS variant. The SoC talks to the ASICs directly over its own serial/SPI lines, runs a custom Linux build (not the OpenWrt found on Antminers), and drives the proprietary btminer mining process. There is no cgminer here and no FPGA bitstream to flash — a genuinely different platform under the same “SHA-256 box” silhouette.

Real-world power and efficiency

The M31S is rated at about 3,220 W for roughly 76 TH/s, which is the ~42.4 J/TH the spec sheet quotes. Treat that as a wall figure for a healthy unit in a cool room: real draw drifts a little higher as intake temperatures rise and the integrated power supply ages, and MicroBT shipped the M31S in several hashrate bins (units in the low-to-mid 70s TH/s were common), so two machines off the same pallet can land a few percent apart. The fans alone account for a meaningful slice of the wall reading once the unit is working hard.

By 2026 standards, 42.4 J/TH is firmly legacy tier. That number decides almost everything about how the M31S should be deployed: at high grid prices it is hard to run for pure profit, but with cheap, curtailed, or behind-the-meter power — or when you actually want the heat — it remains useful hardware.

There is tuning headroom in both directions. Modern mining firmware uses power targeting: you set a watt or hashrate goal and an autotuner calculates the frequency and per-domain voltage at runtime to hit it, rather than loading fixed presets. Because the values are computed live against each unit’s binned silicon, no two M31S units converge on exactly the same operating point. Underclocking trades hashrate for materially better J/TH and a quieter, cooler machine; pushing harder buys throughput at a worse efficiency and more heat. Our ASIC power profiles database shows how that watt-versus-hashrate curve behaves across SHA-256 hardware and where the efficiency sweet spots sit.

Firmware compatibility

Out of the box the M31S runs MicroBT’s stock btminer stack with its own web UI and the Whatsminer JSON-RPC API on TCP port 4028 (current spec V2.0.5). The read side of that API — version, summary, pool, and device queries — is enabled by default, while write operations (pool changes, reboot, frequency/power control) are locked until you explicitly enable them and authenticate with MicroBT’s token-encryption scheme. Stock firmware exposes high-performance, normal, and low-power modes plus a target-power setting, which is enough for most operators.

An aftermarket ecosystem grew up around the M3x platform. Third-party firmware exists that adds autotuning, custom fan curves, and wider power range; and replacement universal control boards — such as the ePIC UMC that D-Central stocks — swap out the stock Allwinner controller entirely for operators who want vendor-independent control with no developer fee. Be honest with yourself about the ceiling, though: none of these options turn a 2020 Samsung 8nm machine into a modern-efficiency miner; they reshape the curve, they do not move generations.

Two firmware realities are worth stating plainly. BraiinsOS does not run on Whatsminer hardware — it is an Antminer-only project — and since BraiinsOS+ is the only firmware that natively speaks Stratum V2, the M31S has no native Stratum V2 path; it mines over Stratum V1. D-Central’s own firmware work (DCENT_OS, in closed beta) is likewise focused on Antminer-class control hardware and does not target this legacy Whatsminer platform, so we will not over-promise it here.

Common faults and troubleshooting

After several years of service, M31S problems cluster into a handful of patterns:

• One hashboard dropping out. With three boards, losing one cuts total hashrate by roughly a third — a 76 TH/s unit suddenly reporting around 50 TH/s is the classic symptom. Causes range from a single failed voltage domain to a broken ribbon connection.
• Over-temperature shutdowns. Dust-clogged heatsinks, dried-out thermal pads, or a hot intake push chip temperatures past the protection threshold and btminer powers the unit down to save the boards.
• Fan errors. The dual fans are monitored for RPM; a seized or out-of-range fan will trip a protection fault and stop mining even if the boards are healthy.
• Power-supply faults. The integrated PSU is a wear item — aging capacitors and a single dirty AC input can produce intermittent “power off” events and unexplained restarts.
• Chip and domain failures. Years of thermal cycling crack solder joints and kill individual ASICs; because voltage is regulated per domain, one shorted chip can take a whole domain offline and show up as reduced board hashrate or elevated hardware errors.

Work symptom-first rather than swapping parts at random. Our ASIC fault finder walks the M31S’s behaviour — dead board, low hashrate, fan fault, no-boot, overheating — back to a likely root cause, and the companion error-code library decodes the specific messages btminer reports so you know whether you are chasing a fan, a sensor, a domain, or the PSU.

Repair and longevity

The M31S was built to be serviced, and it pays to remember that a legacy unit is not a disposable one. D-Central has repaired SHA-256 ASICs in-house since 2016, Whatsminer hardware included, and the M3x generation is well within our bench’s wheelhouse. Board-level work means tracing a failed voltage domain, reflowing or replacing dead Samsung ASICs, recapping or rebuilding the integrated power supply, fitting new fans, and — when the controller itself fails — swapping the Allwinner control board (H6 / CV200-OS SoCs and CB-class boards are available as repair parts rather than throwaway assemblies).

In practice the consumables wear out long before the silicon does: fans, thermal interface material, and PSU capacitors are the usual first casualties. Replace those on a schedule, keep the intake clean, and a 2020 M31S has years of useful life left. If your unit is down or under-performing, our ASIC repair service diagnoses and fixes Whatsminer boards instead of writing them off — that is the whole point of keeping older hardware on the network.

Who the M31S is for

The M31S earns its keep in specific roles rather than as a frontline profit miner:

• Heat recovery. At ~3,220 W it dumps roughly 10,987 BTU/h — the output of a small electric space heater. Ducted into a workshop, garage, greenhouse, or basement, that heat you would have paid for anyway gets converted into hashrate.
• Cheap, curtailed, or behind-the-meter power. Where electricity is near-free, off-peak, or otherwise spilling, a 42.4 J/TH machine is fine; the efficiency penalty only bites when you are paying full retail around the clock.
• Hobby, learning, and solo lottery mining. It is an inexpensive way to run real SHA-256 hardware, learn pool and firmware tooling, or point hashrate at a solo pool for the block-finding lottery.

What it is not is a quiet home appliance. Like all dual-fan industrial ASICs it runs in the ~75 dB(A) range and wants a dedicated, ventilated space — reflected in its low home-mining score — so it belongs in a garage or utility room, not a living room. If near-silent, plug-and-play home mining is the goal, an open-source Bitaxe-class miner is the better tool, and the same shop carries refurbished industrial units and parts for operators scaling up. The M31S is bought on the used and refurbished market today; D-Central both sells reconditioned units and repairs the ones you already own.

Generational context

The M31S sits in MicroBT’s 2020 Samsung 8nm cohort alongside the better-known M30S, and it competed directly with Bitmain’s Antminer S19 from the same year. It was a genuine workhorse of the 2020–2022 cycle. Everything MicroBT shipped afterward widened the efficiency gap: the M50 moved to 5nm in 2022, the M50S++ reached the low-20s J/TH, the M60 series landed near 18 J/TH in 2023–2024, and the 2025 M70 generation runs around 14.5 J/TH.

A 2025 flagship is roughly 2.4× more efficient per terahash than the M31S, but read that for what it is: the gap is in operating cost, not capability. The M31S still hashes valid Bitcoin work exactly as it did in 2020. As long as you value the heat or have very cheap power, a well-maintained M31S remains a sensible way to keep older silicon earning instead of landfilled — which is precisely the kind of hardware we keep alive on the bench.