Autotuning

An ASIC is a specialized chip for mining calculations. Braiins explains that small differences arise during manufacturing, while condition and temperature also affect operation. Mining Firmware, the software controlling the machine, can therefore test the chips individually. Frequency sets the pace of their work; voltage affects the electrical conditions under which they perform it. Some chips handle a larger workload better than others. Think of assigning tasks according to workers’ capabilities instead of requiring everyone to keep the same pace. One machine’s result should not automatically be extended to another. [1]

Andrew might want higher hashrate, meaning more hashing attempts per second, or lower energy consumption for the same work, expressed in J/TH (joules per terahash, the energy needed for a trillion hashing attempts). He may instead need to stay within his facility’s power allowance. These are different goals. Braiins tunes for a selected power level; LuxOS distinguishes requested power from constraints. Autotuning is not a synonym for Overclocking, raising operating frequency: it can also be used with a lower-power setting. Whatever the goal, the PSU, or power supply unit, cooling and electrical circuit limits still apply. [1] [5] [6]

The software first tests what the chips can handle, so intermediate speed is not the final result. Braiins states in its frequently asked questions (FAQ) that duration depends on the model and control board and may extend to several hours. Stable means tuning has settled in that mode, not a promise of permanently trouble-free operation. Andrew also needs to inspect individual hashboards, the boards carrying computation chips: a combined disabled status can reflect a single disabled or faulty board while the others are already stable. A status label needs to be read alongside component-level information. [2]

The Braiins FAQ gives an S19 example with a 3000 W target but only one working hashboard. Allocation based on the expected three boards leaves 1000 W for that board. It does not mean the tuner must redirect all 3000 W to the remaining board. Missing hardware changes what the complete assembly can deliver. Andrew therefore does not automatically treat lower consumption as a tuning fault or try to recover the missing performance regardless of limits. Board, PSU and electrical circuit limits continue to apply. [2]

Braiins distinguishes a traditional saved profile from Continuous Tuning, labelled beta at this editorial review. A saved profile contains selected voltages and frequencies. DPS, or Dynamic Performance Scaling, can switch power levels, but switching alone does not retune the original profile for a different temperature. A continuous tuner instead keeps adjusting settings to conditions. In that mode, thermal safety takes priority over performance targets. The word automatic therefore does not tell Andrew whether software is finding new settings or applying previously saved ones. [3]

LuxOS AutoTuner has its own boundaries. It does not adjust custom profiles with explicitly specified frequency and voltage: the user has chosen that combination. Power targeting, control based on requested consumption, and profile-based tuning cannot operate simultaneously. Manual frequency and voltage commands are also unavailable while power targeting is active. Andrew cannot assume that instructions or experience from another firmware apply here. The active mode and support for the particular hardware model determine the behaviour. [4]

The LuxOS powertargetset command distinguishes power_target, requested consumption, power_limit, the upper bound, and power_min, the regulation floor. The older power parameter sets the target and ceiling together. Changing the target therefore need not change every constraint. It is like a selected speed and a permitted range: related, but different. These names help when reading technical documentation; beginners do not need to run commands to understand the distinction. An active setting is also an instruction to software, not a precise wall-power measurement for the whole machine. [5]

In the calculation below, W denotes power draw in watts and TH/s denotes terahashes per second. Consider a simple illustrative model: 3000 W / 100 TH/s = 30 J/TH. After a change, 2700 W / 90 TH/s = 30 J/TH. Consumption and attempts per second fell in the same proportion, leaving efficiency unchanged. A lower bill may be desirable, but the same computational work still requires the same energy. Andrew compares results under equivalent thermal conditions and measures the same parts of the device each time. He also checks work accepted by the pool, a service that combines miners’ computational work. The machine’s own speed estimate is not enough. Neither Autotuning nor Stable guarantees savings or profit. Measured results, errors and actual operating time decide the benefit. [2] [4] [7]