Smartphone energy consumption
By Joel Lima, E21
In a new paper published at the 2019 IEEE International Symposium on Performance Analysis of Systems and Software, Associate Professor Mark Hempstead worked with collaborators from the University of Buffalo, Google, and the University of Illinois to analyze the differences in the performance and energy consumption of smartphones of the same model.
The researchers performed this study after previously observing that they were unable to reproduce performance results while running a CPU-intensive benchmark test on different smartphones of the same model. The team believed that these differences are due to variations in the underlying transistors in the devices. Variation in transistors of the same model has risen over time due to increasing chip complexity and the accompanying reduction in transistor size.
The researchers found up to a 20% difference in the energy consumption and performance of smartphones that were the same model. They also found that the differences, caused by transistor variations, are exacerbated by temperature differences. Ambient temperature has a big effect on the performance of a phone; benchmark test scores showed a preference towards lower ambient temperature. The score of a bad CPU at a significantly lower temperature significantly out-performs the score of a good CPU using the same benchmark test. Previous studies showed that putting smartphones into refrigerators can improve scores by up to 60%.
Unintended unequal chip creation is an issue that pre-dates the development of smartphones, but the size and operation style of the smartphone poses a unique set of issues. Smartphone manufacturers use voltage binning to overcome transistor differences. All phones of a certain model are made to operate at a constant frequency while their voltage supply varies, to make sure that frequency is met and operation is stable. This means that some phones will have a higher energy demand, and some of this electrical energy is converted to thermal energy, which heats the device.
However, smart devices have a built-in thermal limit and, unlike laptops and desktops, smartphones are too small for most effective temperature regulators. Some devices have been shown to reach their heat limit within 10 seconds, at which point performance slows, leading consumers to believe their devices are malfunctioning.
Hempstead and collaborators developed a comprehensive method for consumers to test variations in their smartphone’s transistors that may be responsible for lower performance than advertised or documented in reviews. The researchers have packaged the method and test into an app now available on the Google Play store.
Authors: Guru Prasad Srinivasa (University of Buffalo), Scott Haseley (Google), Mark Hempstead (Tufts University), and Geoffrey Challen (University of Illinois). Read more soon when Proceedings of the 2019 IEEE International Symposium on Performance Analysis of Systems and Software becomes available online.