When does a faster MCU use more battery energy for the same task?

Compare energy over the whole reporting cycle at the actual supply voltage. For approximately constant voltage, E = V × Σ(I_i × t_i). Include clock startup, sensor acquisition, communication and the time spent waiting for peripherals, not just the arithmetic loop.

As an illustration, 10 mA for 2 ms and 4 mA for 5 ms each consume 20 µC of active charge. Their total cycle energy can still differ because sleep current, regulator efficiency and wakeup costs differ. Measurements on complete boards help expose these hidden contributions.

Meet the real deadline, then evaluate clock and sleep choices with representative inputs. A race-to-sleep strategy helps only when the system can actually enter a lower-power state afterward. If a sensor or interface forces the board to remain awake, faster CPU completion may have little benefit.