Data Center Energy Reduction by Lowering Chip-to-Supply Thermal Resistance

• Published in: InterSociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems pp. 1 - 6
• Main Authors: De Bock, Peter, Bress, Thomas, Lecoustre, Vivien, Gidwani, Ashok, Noyes, Carlos
• Format: Conference Proceeding
• Language: English
• Published: IEEE 30.05.2023
• Subjects: Cooling; data center; Data centers; Data models; energy efficiency; Predictive models; Temperature distribution; Thermal resistance; Thermomechanical processes
• ISSN: 2694-2135
• DOI: 10.1109/ITherm55368.2023.10177591

Data centers are estimated to consume up to 2% of current US electricity production. With increased processor power, rising ambient temperatures, and reduced water availability cooling of data centers is becoming a greater challenge. All electrical power entering a data center must be rejected to the environment as heat. This is typically done via primary and secondary cooling loops. The secondary loop transfers heat from the servers to the facility cooling system. The primary cooling loop rejects heat from the facility cooling system to the environment. The energy cost of rejecting heat to the environment is often high because primary facility coolants are typically kept at 20-30°C, which is relatively close to typical ambient climate conditions. Evaporative cooling or chillers are therefore often used to facilitate heat rejection and maintaining the compute room at acceptable temperatures. As chipsets and electronics can often operate at higher temperatures (typically at 70-90°C), it is here explored how reducing compute room thermal resistance (thermal resistance of the secondary loop) can impact the energy of the cooling plant of the data center as a whole. This paper explores the benefits of lowering the chip-to-coolant thermal resistance and raising the primary loop water supply temperature in data centers and its effect on cooling plant energy and water usage. A simplified Python model of a data center primary and secondary loop was constructed. Hourly data from U.S. Climate Reference Network weather stations in each US ASHRAE climate zone was then used to drive a dispatching algorithm that decided whether dry coolers, evaporative towers or chillers could be used to reject heat to the environment each hour of the year. This dispatching algorithm was optimized to produce results comparable to those published by cooling system manufacturers for a set of six test cases. The optimized model was then used to predict data center energy and water usage in each US climate zone for a wide range of primary loop water supply temperatures. The model results show that lowering the chip-to-coolant thermal resistance allows the water supply temperature to be increased, leading to an increase in the number of hours that dry coolers can be used in every climate zone throughout the year. This in turn leads to an opportunity to reduce cooling energy consumption, operational CO 2 footprint, and water consumption.