Power Measurements of Mobile Computing Platforms

Encadrants : 

Occurrences : 


Nombre d'étudiants minimum: 


Nombre d'étudiants maximum: 


Nombre d'instances : 


Warning : The project will be supervised in English only. The final defense and poster presentation can be prepared in English or French.


Today’s mobile devices (e.g., smart phones, tablets, …) are expected to offer considerable computing power that allows the users to browse the internet, read emails and documents, and even to play 3D video games and watch high-definition videos. At the same time the devices are expected to last for days before being recharged. The power consumption of hardware components (processor cores, graphics cores, WIFI, modem, GPS, …) is thus heavily optimized in order to minimize power consumption. Many modern platforms, for instance, support dynamic voltage and frequency scaling (DVFS) On the software side, however, optimizations targeting the power consumption are relatively new.

Your task is to perform power measurements on a hardware board (TI AM5728 Evaluation board) that is equipped with a typical mobile computing platform (TI Sitara AM5728 consisting of 2 x ARM A15, 2x ARM M4, 2x C66x DSP, 2x SGX544 GPU, …) . The measurement should evaluate the energy consumption of the ARM A15 processor cores for several simple benchmark programs considering different settings for DVFS. Based on the obtained measurements an optimal DVFS setting should be determined for each considered benchmark, if such a setting exists. In addition, the evolution of the energy consumption with regard to the DVFS settings should be studied. The goal of this study is to support or refute the hypothesis that the energy consumption of the benchmarks can be described using a convex function, i.e., a unique optimal setting always exists for each benchmark.


The experiments are performed on the TI AM5728 Evaluation board using measurement instruments of National Instruments (CompactDAQ). The benchmark programs should be adopted from open-source benchmark suites (such as MiBench or Mediabench). The collected data will be processed using National Instruments’ LabVIEW as well as MathWorks’ MatLab software.