Leak Detection and Management

Overview

Leak detection and management is a critical safety and availability capability for liquid-cooled AI rack deployments, where undetected coolant leakage can cause server failures, electrical damage, and unplanned downtime. As liquid cooling transitions from optional to mandatory for high-density GPU racks, OEMs must implement systematic detection and response architectures.

MiTAC's Phase 1 approach uses per-rack Cat 5 water leak detection cables and water trays feeding into the in-rack CDU, with the Rack Management Controller (RMC) monitoring leakage sensor data and sending IPMI commands to gracefully shut down servers within 5 or 30 minutes depending on configured threshold policies. For multi-rack deployments using in-row CDU topology (where there is no per-rack CDU to embed detection logic), MiTAC adds third-party leakage detection hardware (e.g., APC Netbotz 250) at each rack, reporting status to the central RMC.

Phase 2 plans a redesigned lightweight dedicated RMC controller to replace the current general AMD CPU-based server used for rack management, reducing cost and footprint.