Reducing data center energy usage

With thousands of data centers around the world and counting, the energy required to cool servers is immense and will continue to grow. By studying the effects of parameters like airflow and server geometry, Huaitao Zhu and Jiawen Yu worked to develop reliable models to reduce the power consumption required for data center thermal management.

The researchers found the air inlet velocity and temperature are the most critical parameters to control; as expected, higher velocity and lower temperature generally lead to better cooling. However, the layout of the servers — specifically, their height and spacing — also have major impacts, where lower server heights result in improved airflow and temperature control.

“The strong influence of inlet velocity and temperature provides a clear directive for thermal management systems,” said Yu. “Operators can use such models to find the optimal balance between cold aisle supply conditions and energy consumption, avoiding overcooling while ensuring hotspots are prevented.”

Nevertheless, these parameters can be finnicky. For example, airflow velocities that are too high can have diminishing effects. Higher velocities require more energy and can create turbulent flow and hotspots within the servers, in addition to being more likely to carry in dust and other impurities that can block cooling pores and cause equipment failures.

A good balance of air inlet velocity and temperature is 1.5 meters per second at 291.15 Kelvin, which leads to server temperatures within operational requirements at low power consumption.

The authors plan to further refine their models and integrate them into energy optimization algorithms to develop holistic recommendations for data center cooling and energy management.

“We’d like to emphasize that effective thermal management is a multi-scale problem, from the chip level to the room level,” said Yu. “Our work bridges the cabinet and room scales.”

Source: “Numerical investigation of thermal environment in data centers based on fluid flow analysis,” by Huaitao Zhu and Jiawen Yu, International Journal of Fluid Engineering (2026). The article can be accessed at https://doi.org/10.1063/5.0288940 .