Energy-efficient task scheduling algorithms in hybrid CPU-GPU systems

This article presents a systematic analysis of energy-efficient task scheduling algorithms in hybrid CPU–GPU systems, where heterogeneous computing resources exhibit fundamentally different performance and energy characteristics. Such systems are widely used in high-performance computing, data centers, and AI-driven workloads, where increasing computational demand and energy constraints limit the effectiveness of scheduling strategies focused solely on execution time. The study is conducted as a review-and-analytical synthesis of peer-reviewed publications published between 2022 and 2025, without quantitative aggregation of results due to heterogeneity of experimental setups, metrics, and energy models. Particular attention is paid to the role of optimization metrics, quality of input energy data, and integration of power management mechanisms in shaping the observed effectiveness of scheduling algorithms. The analysis shows that reported improvements strongly depend on the choice and aggregation level of performance and energy metrics, as well as on the accuracy of task energy characterization, rather than on the algorithmic class alone. It is demonstrated that single-objective formulations fail to capture the behavior of hybrid CPU–GPU systems, while composite time–energy criteria and explicit device-selection policies provide a more consistent basis for evaluation. The study establishes that meaningful comparison of energy-efficient scheduling approaches requires a clear separation between architectural scheduling frameworks and concrete algorithms, along with explicit fixation of the energy management context. The article is intended for researchers and practitioners working on scheduling, power-aware computing, and heterogeneous system design in high-performance and data-intensive environments.