Here is the abstract you requested from the Thermal_2012 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.
|Investigation of Passive Two-Phase Cooling for High Performance Computers – Preliminary Results from Test Platform|
|Keywords: Liquid Immersion, Two-Phase, Test Platform|
|High Performance Computing (HPC) systems are pushing the limits of current thermal management methods. HPC system loads have reached the point where air cooling is often infeasible, requiring advanced cooling methods, such as liquid cooling, which can be costly and complicated to implement. An alternative approach is the use of passive two-phase immersion cooling techniques using high performance dielectric fluids. (Note that this is different from the single phase immersion cooling currently employed by Green Revolution or used by Cray Research in the past.) Previous work shown at this conference has demonstrated that electronics can be cooled efficiently using fluoroketones, in particular, Novec 649 from 3M. Results of preliminary experiments indicate that high-speed links (beyond 15 GHz) and certain optical connector configurations can be immersed in Novec 649 without affecting performance1. Mayo Clinic is also conducting an extended run experiment using a Dell Optiplex 755 PC equipped with a solid state hard drive. This system has been running an application while immersed in Novec 649 for almost two years without failure. In addition to experimental data, calculations based on the thermal properties of the Novec 649, show heat removal rates of 4 kW per liter of dielectric fluid2. This heat transfer rate points to greater packaging densities. In addition, because all of the electronics are immersed, the thermal environment is more uniform, which may increase component reliability. This technology looks promising at a small-scale; however, at large-scale two major questions remain to be answered: Can this technology scale to meet the performance requirements of HPC systems, which typically start at 50 kW/rack? Can this be accomplished in a cost effective manner? To answer these questions requires first identifying and then investigating the potential road blocks to implementing a passive two-phase immersion cooling system: enclosure design, system availability, system performance, maintenance, and cost of ownership. To this end, Mayo has set up a fully instrumented immersion cooling test platform which is capable of producing up to 75 kW of power. The effort of designing and installing the test platform has provided insight into the facilities infrastructure and enclosure design requirements needed to implement immersion cooling for a system. At this time, the test platform has been used to evaluate the scaling performance of two different condenser designs. This presentation covers the following: brief background on immersion cooling including motivation for pursuing this technology; overview of primary implementation issues; description of test platform; summary of test platform installation requirements and considerations; and discussion of the condenser experiments. The talk concludes with a summary of open issues.|
|Stephen C. Polzer,