Peak power for a Power Manager or power domain, is an indication that several devices are in the Active (full power) state at the same time, such that the peak, or instantaneous power is at a Maximum. In a clocked system, this may indicate that several devices are active at the same time for a particular task or thread.
If one examines the devices contributing to the peak power, then one can get an idea of which device’s activity might be shifted to another clock, or multiple clocks. The goal is to reduce the average power by shifting certain device activities to an adjacent timeframe. One could argue that leaving the original peak power, average power is the same, regardless of execution in time. However, the peak power can affect the Thermal Design Power (TDP), such that if the TDP maximum is exceeded for even a short time, then the reliability of the affected devices may be reduced in the long run.
It is also true that for a mobile, battery-based Digital System, that the battery life is reduced with volatile power consumption over time. This is a second reason to consider smoothing out overall peak and average power consumption. This applies to mobile smart phones to electric vehicles. The one disadvantage of shifting device per activity is that the overall task or thread latency might increase somewhat. The best situation is to access memory, while the HW accelerator is at a different time from memory access. If the activities are pipelined, then spreading the power consumption results in the best reliability. There are power tradeoffs when altering peak or average power over time. Next, will consider consolidating power consumption to be sequential, using the above techniques, to reduce the time certain devices must remain in the Active state.