An ASHRAE podcast recently delved into a critical evolution within data centres: the increasing necessity of liquid cooling. Host Justin Seter guided a panel of industry experts – David Quirk, Dustin Demetriou and Tom Davidson – through the intricacies of this technology, driven by the insatiable demands of artificial intelligence (AI) and high-performance Graphics Processing Unit (GPU) applications. This is Part 9 of a nine-part series.
A business-specific approach to tracking performance per watt might be more valuable than a universal metric. Image by Joshua Sortino/Unsplash
The conversation then shifted to the unique design considerations for high-density liquid cooling applications. Quirk emphasised that “everything is new” in this rapidly evolving landscape, including equipment, vendors, design criteria, coolants and commissioning processes. He highlighted the influx of new players, such as CDU manufacturers, for whom standardised testing methods are still lacking. Quirk also noted the emergence of thermal energy storage and the increased need for UPS-backed pumps to ensure continuous flow in liquid cooling systems. Advanced modeling techniques, combining hydraulic system analysis with computational fluid dynamics (CFD), are becoming essential to understand the dynamic interplay between air-cooled and liquid-cooled components under both steady-state and transient conditions, given their differing thermal time constants.
Demetriou underscored the critical need for access to detailed physical planning data for liquid-cooled servers, such as pressure drop characteristics when adding devices to the loop. To address this, TC 9.9 recently released a liquid cooling server thermal template, analogous to the existing air-cooled server template, providing crucial information like the percentage of heat captured by the liquid loop versus air, pressure drop on both the liquid and air sides, and maximum operating pressure. This is particularly important given the delicate nature of cold plates and the potentially high pressures within the cooling loops.
Finally, Seter posed a crucial question regarding the load profiles that these new high-density liquid-cooled systems are being designed for, and whether these profiles are even well-understood at the design stage.
Quirk responded that the industry is currently casting a wide net, often attempting to design facilities that can simultaneously accommodate 100% air-cooled environments alongside areas with a high percentage of liquid-cooled racks at five times the power density. This leads to complex overlays of piping networks on the facility water side, provisioned to support future CDU deployments. He described this as “partial future-proofing”, with significant guesswork involved in anticipating the timelines for large-scale liquid cooling adoption, leading to potentially costly over-provisioning.
The podcast concluded with the panelists offering guidance on how to stay updated on this rapidly evolving field. ASHRAE, with its comprehensive data centre publications and the ongoing work of TC 9.9, SPC 127, and SSTC 9.4, remains a primary resource. The Open Compute Project (OCP) also regularly publishes relevant white papers. Industry conferences and podcasts like the ASHRAE Journal podcast itself were highlighted as valuable sources of information. Demetriou further plugged the TC 9.9 DataCom Encyclopedia, now an online platform with quarterly updates, as a key resource for the latest developments in data centre cooling.
