Integrated Solutions for Data Centers

Jason MacDowell – GE Vernova’s Consulting Services

AI data center workloads are rapidly growing and have special electrical power consumption characteristics that need to be considered. Machine learning applications, particularly training ML models, pose a rapid load cycling demand from power generation and the grid. This rapid load cycling must be mitigated to avoid adverse power quality and torsional stress imposed on turbine-generator drivetrains. Additionally, IT loads may impose negative electrical damping that also needs to be mitigated. Third, the power architecture needs to be resilient and ride through electrical disturbances, accounting for the behavior of the UPS, data center load, generation and substation protection. Finally, all of this needs to be considered while delivering a robust system designed with high reliability performance expected by hyperscalers and end users. Power system design architectures will be presented that address these various needs and challenges for data centers powering AI workloads.