Modeling and optimization of power consumption for multi-core processors: A thermodynamic perspective

As artificial intelligence advances intensify demands for efficient computing, this article addresses the fundamental limitations of traditional processor power models rooted in empirical frequency/voltage parameters. A stochastic thermodynamics-based power consumption model for multi-core processors is proposed in this article, revealing an exponential temperature-dependent power consumption escalation through thermodynamic analysis. Based on the relationship between processor power consumption and temperature, a spatio-temporal optimization algorithm is designed to reduce the power consumption of multi-core processors, which achieved maximum 22.2% power consumption reduction compared with unoptimized processors. Furthermore, temperature-induced voltage offsets in optimized processors exhibit an overall reduction of up to 52.45% compared with processors without optimization. Algorithmic cost analysis demonstrates superiority over traditional load balancing and layout optimization when algorithm cost factors fall below 0.016 and 0.17, respectively. • Modeling of processor power consumption based on Stochastic thermodynamics. • Higher temperatures exponentially increase multi-core processor power consumption. • New algorithm optimizes core layout and runtime to reduce power consumption. • Optimization reduces processor power consumption by up to 22.2% in simulations.