Understanding the Sources of Power Consumption in Mobile SoCs

In this paper we propose a fine-grain blind leakage and dynamic power identification technique, and we use it to analyze the dynamic and leakage power of mobile SoCs at the core level. We also introduce a new experimental methodology to apply blind power identification for heterogeneous SoCs, including a novel initialization for the algorithm that enhances its power estimation accuracy. We shed light on power usage using real life applications, showing how power is divided among the big, the LITTLE cores and the GPU. Our results show that for some applications, the GPU can have the highest power consumption, and that the LITTLE cluster can have large values of leakage power. We also elucidate the trade-offs between power consumption and performance of the big cluster versus the little cluster of the SoC at different frequencies. We also show how the power is affected by CPU and skin thermal throttling.