As data centers grow exponentially in size and number worldwide, it appears evident that more efficient and reliable cooling solutions are needed to reduce energy consumption and noise level significantly. Thermosyphon-based systems have been proven to respect both these goals, and their inherent natural circulation makes them very attractive for heat rejection of individual servers and entire server racks. In this study, the thermosyphon concept is applied as a retrofit cooling system on an actual server rack. The cooling system is composed of two stages: (i) server-level thermosyphons for the direct cooling of the two Central Processing Units (CPUs) and (ii) a rack-level thermosyphon for the cooling of the condenser side of the server-level thermosyphons. Nearly zero-Global Warming Potential (GWP) refrigerants, R1234ze(E) and R1234yf, are used as the working test fluids for the server-level thermosyphons, while for now R1234ze(E) is the only fluid used for the rack-level thermosyphon. A server-level thermosyphon is first tested when cooled directly by water at 20 degrees C, without a rack-level thermosyphon, to define its thermal performance. Then, the rack-level thermosyphon is installed in such a way that it serves the subcooled liquid at the server-level thermosyphons in a parallel configuration. The condenser of the rack-level thermosyphon is cooled by water entering at 20 degrees C.

Experimental results show that the maximum CPU's temperatures are stable at about 40 - 45 degrees C when they work at their maximum load, which is 25 to 40 K less than the commercial air-cooling solution before the conversion from air to thermosyphon cooling. Fan power consumption is reduced by 100 W, and the rack-thermosyphon conveys all the heat rejected directly to the chilled water, requiring no additional pumps or fans. This demonstrator and its results show the potential of thermosyphon cooling technology to achieve higher system performance. In particular, highly performing CPUs can be safely used in data centers as there is not the heat transfer limitation of air-cooling. The proposed approach would allow significant efficiencies to be made in the facility air-handling systems, ultimately leading to operational cost savings. In addition, maintenance is reduced significantly as no moving parts are required for gravity-driven thermosyphon. The heat rejection occurs directly to the chilled water, without the need to pass through air handlers and coils, reducing the mean temperature difference and increasing the Coefficient of Performance (COP) of the chiller significantly (the COP can increase by a factor of 2 to 5, depending on the cases).