Exploring the Energy Consumption of Highly Parallel Software on Windows

Studenteropgave: Kandidatspeciale og HD afgangsprojekt

  • Jamie Jonas Baldwin Pedersen
  • Mads Hjuler Kusk
  • Jeppe Jon Holt
4. semester, Datalogi, Kandidat (Kandidatuddannelse)
With the evolution of CPUs over the last years, increasing the number of cores has become the norm resulting in additional resources for software to utilize. Through four research questions, this research investigates the performance gains obtained from the additional processing power and the impact of the P- and E-cores on parallel software. The experiments are conducted on two computers, where the analysis is made based on energy consumption and execution time on a per-core basis and on an increasing number of cores. The experiments are conducted primarily on Windows, where Intel's Running Average Power Limit is unavailable, with Linux as a reference point. The energy consumption on Windows is measured on the best performing measuring instrument, found through a set of experiments when running both micro- and macrobenchmarks, where measurements are compared against a ground truth.
Udgivelsesdato9 jun. 2023
Antal sider15
ID: 531223839