Antminer Underclock & Tuning Envelopes

Search for "Antminer underclock settings" and most results hand you a frequency and a voltage. For S19- and S21-generation hardware that answer is fabricated: the firmware's preset system stores watt-to-hashrate operating points only. You choose a power target; the autotuner then benchmarks your specific hashboards and derives per-chip frequency and per-domain voltage to hit that target. Two units of the same model land on different settings — which is why no honest page can print them in advance.

What CAN be published is the envelope: how low the catalog goes for each model, where efficiency peaks, and how much efficiency you give up at higher targets. Three patterns emerge from the data below:

• Flat curves (S19k Pro, S21 Hydro, S21+): efficiency barely moves across the whole envelope — the S19k Pro holds 25.6–26.0 J/TH from 2,050 W all the way to 4,860 W. On these models your watt target is purely a heat, noise and circuit decision.
• Rising curves (S19, S19j Pro, S19j Pro+): best J/TH sits at or near the floor — the lowest watt target your hashrate goal tolerates is also the most efficient point.
• Knee curves (S21 XP Hydro): efficiency holds flat up to a knee, then degrades — run at or below the knee unless you need the extra hashrate.

Legacy S17-generation presets and S9-era firmware work differently (frequency-based); their pages note it explicitly. For overclocking headroom — the opposite direction — see the ASIC overclock headroom dataset; for step-by-step firmware instructions, the Antminer undervolting guide.

* 120V fit = the envelope FLOOR is at or under the NEC 80% continuous-load budget (1,440 W on a 15 A circuit, 1,920 W on 20 A). It is a circuit-budget check only: stock Bitmain APW-series PSUs require 200–240 V input, so 120 V operation additionally needs a PSU swap or rework. Sweet spot = the contiguous watt band within 5% of the model's best efficiency.

Bands summarize only the models with published envelopes above (best J/TH per mode, min–max per chip). They are NOT full-generation claims — models without extraction data (see below) do not contribute. Chip assignments follow the D-Central chip canon and the A-graded overclock-headroom dataset; no new chip attributions are made here.

The most-searched air-cooled flagships are exactly where the public record is weakest — and where we refuse to fake it. No extraction set exists for these units, so this page publishes no numbers for them rather than dressing up a different SKU's data:

When a set for one of these units clears source-integrity checks it will appear here with its status; until then the manufacturer stock rating (see each model's spec page) is the only defensible number.