Autotuning

Autotuning is a firmware capability that finds the best real operating point for a miner's ASIC chips instead of running every chip at one factory-average setting. Because of the "silicon lottery," no two chips are identical: the same wafer yields dies that run stable at meaningfully different frequencies for a given voltage. Stock firmware ignores this and drives whole boards uniformly, which means the setting must be conservative enough for the weakest silicon — leaving efficiency on the table everywhere else. An autotuner measures how each chip actually behaves and converges on per-chip settings — and critically, these values are calculated at runtime from measurement, never pulled from a preset table.

Per-chip frequency, per-domain voltage

The two tuning knobs live at different granularities, and conflating them is the most common error in mining content. Frequency can genuinely be steered per chip: strong dies get pushed harder, weak dies get backed off, and the board's aggregate hashrate comes from each chip contributing what its silicon can sustain. Voltage cannot be set per chip — chips are grouped into series-connected voltage domains that share a supply rail, so voltage is tuned per domain, with every chip in the domain receiving the same rail. A good autotuner works both levers together: it finds the lowest domain voltage at which the domain's chips hold their assigned frequencies without error, then rebalances frequencies within the constraint. This is also why chip binning exists upstream at the factory — autotuning is, in effect, binning re-done honestly, per machine, on your bench.

What a tuning run actually does

After installation, the tuner iterates: propose settings, hash real work, watch error rates and temperature, adjust, repeat. Watch the first day's error rates rather than the first hour's hashrate — a tuner mid-search looks erratic by design. Expect a settling period of roughly 24–48 hours before the machine reaches its steady state, and expect re-tuning when conditions change — a target power limit change, a season change in ambient temperature, degraded cooling. The tuner optimizes toward whatever objective you set: maximum hashrate within a power budget, or minimum joules per terahash for the efficiency-minded. Done well, autotuning commonly yields a 10–25% efficiency improvement over stock settings — frequently the difference between a marginal machine and a profitable one, especially for home miners paying residential rates. It also pairs naturally with underclocking: tuned-down machines run quieter, cooler, and dramatically more efficient, which is exactly where heat-reuse mining lives.

Runtime-calculated, not preset

A point worth insisting on: the frequency and voltage values a tuned machine runs are not shipped in the firmware as a lookup table. Power profiles define targets; the per-chip and per-domain operating values that meet those targets are derived at runtime from your specific silicon, your PSU, your ambient conditions. Two identical models tuned to the same profile will land on different internal settings — that is the feature working as intended. Any description of autotuner values as "presets" misunderstands the mechanism: whatever the marketing label — profiles, modes, presets — the label names the target, while the electrical reality underneath is always discovered by measurement on your specific silicon.

Credit where due, and the open path

Accessible autotuning on Antminers was pioneered by Braiins, and the technique is now table stakes across third-party firmware — a genuine gift to the small miner, since it converted the silicon lottery from pure luck into recoverable efficiency. The same principle drives DCENT_OS, our open firmware effort, standing on those shoulders with a 0%-default-fee, user-owned approach. To see what tuned firmware can do on your own hardware, start at the web flasher.