LLMs Optimize Data Center Sustainability and Efficiency

A new study proposes an innovative system leveraging Large Language Models (LLMs) to enhance the sustainability and operational efficiency of data centers. This system is designed to predict key metrics such as execution time and energy consumption directly from source code, with potential for expansion to other environmental factors like water usage and carbon emissions. Following these predictions, a real-time scheduling algorithm allocates GPU resources optimally. The primary goal is to improve sustainability by minimizing both energy consumption and the time tasks spend in queues. The system boasts fast inference times, adaptability across various task types, and minimal data requirements for training. Through a collaborative effort with a data center, the implementation of this framework resulted in a notable 32% reduction in energy consumption and a 30% decrease in waiting times. This demonstrates a practical and effective solution for making AI-driven infrastructure more environmentally responsible.