Abstract

Direct to chip liquid-cooling technique has been widely implemented for the cooling of processors with high thermal design power. To further enhance the efficiency of liquid cooling, ongoing research focuses on the optimizations at the cold plate level or by changing the flow configurations. But in all cases, the coolant which is pumped across the rack is pumped at a constant flow rate irrespective of the workload utilization at the individual server, resulting in excess pumping of coolant. A practical approach is to dynamically vary the flow rate to each server as per server workload utilization. In this study, transient analysis is performed by varying the flow rate across individual servers at rack level using CFD. A CFD model mimicking four servers placed at different heights on a standard 42U rack is developed. The flow variation through each of the servers is done using a damper arrangement representing a flow control device. A controller is integrated to automate the process of opening and closing of FCD to vary the flow based on the average outlet temperature from each server. A baseline simulation with all servers running at maximum power dissipation with a constant coolant flow rate is compared with cases where the coolant flow rate is controlled dynamically for varying thermal design power (TDP). The results shown analyze the impact of the dynamic response of the flow control device on transient thermal and hydraulic characteristics across the rack is done.

This content is only available via PDF.
You do not currently have access to this content.