Scaling a National Data Center for Massive AI Workloads

Scenario

A national data center, tasked with supporting the nation's digital backbone, needed to deploy high-performance AI platforms like the NVIDIA Vera Rubin, Blackwell, NVIDIA HGX, and AMD MI350 series on a massive scale. This large-scale deployment involved not only substantial investment but also required professional tech support to ensure the stability for continuity of national services. Along with stability, scalability and high-efficiency cooling was a top priority.

Pain points

1. How to create cooling capacity sufficient to support such high-density computing and prevent GPU downtime from overheating

2. How to effectively monitor the stability of hundreds of water loops be monitored, with data-based accuracy

ASUS solution

In the lab, ASUS addressed these concerns with a live demonstration. We showed a single CDU stably supporting a XA GB721-E2 (NVIDIA GB300 NVL72) with a massive 1,350kW cooling capacity. The system was also shown to maintain a constant inlet water temperature of 27° ± 0.3°C, showcasing exceptional thermal control. Using the ASUS Control Center Data Center Edition platform, the client could monitor the inlet/outlet temperature, fan speed, voltage, and real-time power consumption of every rack, ensuring all parameters remained within safe limits.

Outcome

The 1,350kW cooling curve data not only proved the reliability of the ASUS solution but also became a critical reference for client expansion plans, providing confidence while saving the cost of building a test facility. We also exported the monitoring logic and alert thresholds, helping the client rapidly establish their own systems and maximize deployment efficiency, ensuring the national data center could expand its AI capabilities quickly and robustly.

FAQ