Liquid Cooled AI Workload Case Study

This case study details the collaboration between VIRTUS and Customer A to address the challenges posed by a multi-megawatt AI workload that exceeded traditional air-cooling capabilities. The document outlines the specific requirements for the deployment, including the need for high power density, rapid cooling response, and proximity of networking IT equipment to high-performance computing (HPC) workloads. VIRTUS implemented two cooling technologies to accommodate different styles of IT equipment, utilizing a direct-to-chip liquid cooling solution for the HPC deployment. The study describes the structural adaptations made to support the weight of the deployment and the commitment to using 100% renewable energy sources. It also highlights the commissioning process, which involved innovative solutions to simulate cooling demand and ensure operational efficiency. The case study concludes with the benefits gained from this deployment, including valuable experience for future projects and a low operational Power Usage Effectiveness (PUE) of less than 1.1.