ASSESSMENT OF RELIABILITY ENHANCEMENT IN HIGH-POWER CPUs AND GPUs USING DYNAMIC DIRECT-TO-CHIP LIQUID COOLING
Pardeep Shahi, Amith Mathew, Akash Akash, Satyam Saini, Pratik Bansode, Rajesh Kasukurthy, Dereje Agonafer
Abstract
Thermal management of high-performance computing servers is becoming a prevalent challenge for the data center cooling industry due to increasing power densities at the server level and data center level. Efficient heat dissipation is also directly related to electronic package reliability. Improved cooling technologies like direct-to-chip liquid cooling can address the rising cooling demands due to the higher thermal performance of water-based coolants. A methodology to further enhance the efficiency of direct liquid cooling is experimentally investigated using the concept of dynamic cooling. A Flow Control Device (FCD) was developed to regulate the flow rate to four custom-made Thermal Test Vehicles using ceramic heaters. The TTV assembly was placed at four different levels in a standard 19-inch ITE (Information Technology Equipment) rack in test fixtures mounted with cold plates. The flow regulation to each of the TTVs was done based on the power dissipated by each TTV. The power dissipation of each TTV was varied for various non-homogeneous power distribution values in the entire rack. The influence of coolant inlet temperature and flow rate on the TTV temperature and rack pressure drops was analyzed. The results indicated more uniform temperatures on the TTVs and a reduction in the maximum temperature on the TTV with maximum power. The impact of the temperature uniformity on package level reliability was also analyzed by comparing the results obtained with the published literature.