GAAMP: Automatic Thread Count/Affinity and DVFS Tuning for Asymmetric Multicores
Marcelo K. Moori, Arthur F. Lorenzon, Hiago Mayk G. de A. Rocha, Antonio Carlos Schneider Beck · 2025
Asymmetric multicore processors (AMPs), combining high-performance and energy-efficient cores, are gaining popularity in desktop and server systems like Intel’s Alder Lake and Apple’s M2 for energy savings. To fully utilize their resources, thread count, mapping, and CPU frequency need to be tuned in more complex scenarios than homogeneous systems. We propose GAAMP, a genetic algorithm that automatically determines the optimal configuration for high-performance applications. Using nine NAS benchmark applications on an Intel Alder Lake processor, GAAMP reduces the Energy Delay-Product by 52.56% (up to 78.94%) compared to the default Linux configuration and outperforms fourteen other strategies by 53.65% (up to 98.88%).