Analytical modeling of energy efficiency in heterogeneous processors

• We study how performance per watt is affected in CPU–GPU processors.
• We present extensions to Amdahl’s Law by accounting for energy limitations.
• We analyze three processing modes available for heterogeneous computing.
• We present a new processing mode called simultaneous asymmetric.

Energy efficiency is increasingly critical for multi-core processors and many-core accelerators. In order to increase energy efficiency, chip manufacturers are developing heterogeneous CMP chips.We present analytical models based on an energy consumption metric to analyze the different performance gains and energy consumption of various architectural design choices for hybrid CPU–GPU chips. We also analyzed the power consumption implications of different processing modes and various chip configurations. The analysis shows clearly that greater parallelism is the most important factor affecting energy saving.

Figure optionsDownload as PowerPoint slide