Here is the abstract you requested from the Thermal_2011 technical program page. This is the original abstract submitted by the author. Any changes to the technical content of the final manuscript published by IMAPS or the presentation that is given during the event is done by the author, not IMAPS.
|Passive 2-Phase Immersion of Datacom Equipment - A Comparison to Water Cooling|
|Keywords: Immersion, Cooling, Liquid|
|2-phase immersion cooling is compared to pumped water cooling with a focus on system complexity, thermal efficiency, and power density. The water-cooled systems discussed are the IBM P575 and Aquasar supercomputers. Immersion cooling data comes from several recent experiments conducted with heaters and with IBM, NVidia, Intel and AMD chipsets. CPU simulators in a 4kW/liter array are used to simulate extreme node-level power densities. Like the water-cooled system, an immersion-cooled system is assumed to deposit its heat to facility water via a secondary heat exchanger. Previously published condensation data for condenser power densities of 2–4kW/liter and water flows of 0.4–4 liter/min-kW permit a system level power density and thermal efficiency comparison of the water-cooled system and its hypothetical immersion-cooled equivalent. The final junction-to-water thermal resistance based on the facility water temperature exiting the system is lower for an immersion cooled system Designing an open-bath 2-phase immersion-cooled system is simple compared with a pumped water system but it requires some knowledge of the behavior of two phase processes, knowledge that is often counterintuitive and not typically taught in university. The essential features are demonstrated via a 4 node, 8 socket server cluster running 48AMD Opteron cores running in a hydrofluoroether liquid. The fluid losses associated with “hot swapping” of nodes are quantified.|
|Phillip Tuma, Advanced Application Development Speciaist
St. Paul, MN