AI-Powered Optimization of Multi-Core Processor Performance Using Reinforcement Learning

Abstract

The increasing complexity of modern multi-core processors necessitates advanced optimization strategies to enhance performance, power efficiency, and task scheduling. Traditional heuristic-based methods struggle to adapt to dynamic workloads, leading to suboptimal core utilization. This study proposes an AI-powered reinforcement learning (RL) framework to optimize multi-core processor performance by dynamically adjusting task scheduling, power allocation, and cache management. The proposed Deep Q-Network (DQN)-based scheduler continuously learns and adapts to varying computational demands, ensuring energy-efficient task execution while maximizing throughput. Experimental results on benchmark datasets demonstrate a 23.4% improvement in execution time, a 19.6% reduction in power consumption, and a 31.2% increase in overall system efficiency compared to traditional scheduling algorithms. The findings suggest that reinforcement learning can provide an adaptive, self-optimizing mechanism for managing multi-core architectures in real-time. This work contributes to the nextgeneration intelligent computing paradigm, paving the way for AI-driven, energy-efficient, and performance-optimized processor architectures.