This article explains and compares three heat rejection architectures applied to closed water circuit cooling systems for data centers: air condensing chiller, water condensing chiller and dry cooler with adiabatic support. The analysis uses an IT load of 1 MW, 100 racks of 10 kW, a data room maintained at 22 °C at the height of two thirds of the rack and relative humidity of 55%, considering the climate reality of Santana de Parnaíba, in the Metropolitan Region of São Paulo.

The literature synthesis indicates a recurring tension between energy and water: water chillers tend to reduce electrical consumption, but shift part of the impact to evaporative consumption; air chillers simplify implementation and practically eliminate process water, but increase annual electrical consumption; Adiabatic dry coolers reduce compressors and may have low water consumption, but they do not guarantee low enough cold water at all hot and humid times for a room maintained at 22°C without mechanical support.

For the adopted comparative basis, annual electricity consumption is estimated at approximately 3,329 MWh for air chiller (PUE 1.38), 2,190 MWh for water chiller (PUE 1.25) and 788 to 1,840 MWh for adiabatic dry cooler (PUE 1.09–1.21, depending on support needs mechanic).

PUE values do not include UPS and electrical distribution; The actual facility PUE is typically 0.05 to 0.10 higher. Annual water consumption is estimated at less than 20 m³ in the air chiller, approximately 22,000 m³ in the water tower chiller and 5,000 to 8,000 m³ in the adiabatic dry cooler. This edition incorporates six improvements over the original version: (a) explicit PUE for all scenarios; (b) note on efficiency conservatism adopted for the air chiller compared to modern equipment; (c) discussion on partial load analysis (bin method); (d) impact of redundancy (N+1 and 2N) on real PUE and CAPEX; (e) quality of water supply in Santana de Parnaíba and its effect on real WUE; (f) noise requirements and applicable acoustic legislation (CETESB/CONAMA).