Authors: Jeff Autor (Vertiv, Redfish), David Martinez (Sandia National Laboratories, Energy Efficient High Performance Computing Working Group (EE HPC WG)), Tiziano Belotti (Swiss National Supercomputing Centre (CSCS)), Steven Martin (HPE), Karsten Kutzer (Lenovo), Chris DePrater (Lawrence Livermore National Laboratory (LLNL)), David Grant (Oak Ridge National Laboratory (ORNL), Energy Efficient High Performance Computing Working Group (EE HPC WG)), Michael Ott (Leibniz Supercomputing Centre (LRZ))
Abstract: As direct liquid cooling scales, the interface between IT systems and facility mechanical infrastructure becomes more critical. This BoF explores four challenges. First, high-temperature cooling (heat-reuse, IT thermal margins) and dew point (prevent condensation). Second, warm-water fluid quality (filtration and added chemicals to prevent corrosion, biological growth, and scaling). Third, temperature and fluid dynamics with rapid workload power swings (volumetric expansion, pump cavitation, facility response characteristics). Fourth, integrating telemetry into IT systems and building automation (predict thermal excursions, reduce trapped capacity). This session is intended for facility engineers, system integrators, operations, administrators, architects, and researchers in energy-efficiency and sustainability.
Long Description: As high-density HPC and AI systems move to liquid cooling, the boundary between facility infrastructure and IT systems requires tighter coordination. This BoF examines the physical and software interfaces and the decisions required to deploy and operate large-scale, high-density liquid-cooled systems.
The session addresses four interconnected topics. First, participants will examine the trade-offs of high-temperature liquid cooling. Facility engineers seek elevated water temperatures to eliminate mechanical chillers and enable heat reuse, while system integrators must manage tight thermal margins, protect uncooled peripheral components, and integrated dew-point control to prevent condensation. Second, the session explores warm-water fluid quality. Operating loops between 30 and 50 degrees Celsius accelerates biological growth, mineral scaling, and galvanic corrosion. The facility must deploy filtration and chemical dosing to maintain the fluid parameters required by IT systems. Third, participants will examine temperature and fluid dynamics and hydraulic stability. Rapid workload-driven power swings trigger thermal expansion and pressure drops. The facility must implement dynamic flow regulation to prevent pump cavitation and water hammer. Fourth, the discussion covers telemetry integration. IT power data must integrate with traditional building automation protocols. This allows the facility to predict thermal excursions and adjust flow before hardware overheats, which allows for reducing trapped capacity caused by worst-case provisioning.
Below are some example questions:
1. At inlet temperatures exceeding 45°C, what are the practical thermal limits of uncooled components, such as voltage regulators, and other peripheral devices located within server cabinets?
2. During rapid AI workload spikes, how should facilities balance fast valve response and flow control against the risks of pressure surges, flash boiling, and pump cavitation?
3. How quickly can high-frequency telemetry from hardware interfaces, such as Redfish power data, be securely translated into actionable Building Automation System (BAS) controls using protocols such as Modbus or BACnet?
4. What water quality requirements and operational boundaries do system integrators place on facilities regarding filtration, chemistry control, and automated dosing?
5. How can predictive thermal telemetry safely reduce conservative “wall-plate” engineering assumptions and reclaim trapped capacity?
6. If supply temperatures must be reduced during extreme conditions, how can facilities and IT systems dynamically coordinate dew-point control to prevent condensation and equipment damage?
Next-generation systems drive higher rack densities and facility power demands. This session provides insight for facility engineers, system integrators, system administrators, infrastructure architects, procurement teams, and operators designing, purchasing, commissioning, and operating liquid-cooled environments.
This BoF builds on a series of SC community discussions on liquid cooling infrastructure, controls, and commissioning held between SC13 and SC25. These sessions consistently attract participants from supercomputing centers, system vendors, engineering firms, and research organizations. The attendance reflects a continued need for practical guidance in this area.
The focus on liquid cooling infrastructure remains similar to prior years, but the format shifts from site case studies to discussion between facilities, system integrators and the audience. To maximize interaction, the session uses a facilitated dais format featuring practitioners from supercomputing centers and representatives from system integrators. The goal is to negotiate shared expectations regarding interfaces, responsibilities, and practices to support high-density infrastructure.