Traditional data centers used fans to move heat away from server racks, while cooling systems circulated chilled water through the facility to lower air temperatures. The heated water was then sent to a cooling tower, where the heat was released into the atmosphere. Because some of that water evaporates during the process, evaporative cooling can consume significant amounts of water, says Eric Massanet, a data center sustainability researcher at the University of California, Santa Barbara.
AI data centers face a greater cooling challenge. Their energy-intensive processors generate far more heat than conventional computing hardware, making it difficult to cool servers with air alone. As a result, companies including Amazon, Microsoft, and Google are adopting a more efficient approach known as liquid cooling.
Liquid cooling uses a network of pipes filled with water, sometimes combined with temperature-regulating chemicals, to cool processors directly. Unlike traditional evaporative cooling systems, many liquid-cooling designs operate in a closed loop, meaning the fluid can be reused without continuous water loss. The heated liquid can be cooled with outdoor air when conditions are favorable or with an energy-consuming refrigeration system when temperatures are higher.
During extremely hot or humid weather, data centers may still need additional water-based cooling systems. One approach is to atomize water into the air surrounding the cooling pipes, allowing evaporation to remove excess heat. Vaibhav Bahadur, a mechanical engineer at the University of Texas at Austin, says many AI data centers in Texas use this type of water-based cooling primarily during the hottest summer months.
Bahadur and his colleagues recently estimated that Texas data centers currently consume less than 1 percent of the state’s total water supply, although the sector continues to expand. — Data centers could account for between 3 and 9 percent of the state’s water use by 2040. However, Bahadur expects that share to decline as cooling technology improves. “Data centers are becoming much more efficient in their water usage,” he says.
Recent advances could also reduce the need for water-based cooling in AI data centers. Josh Parker, head of sustainability at NVIDIA, says newer cooling technologies can work effectively even in hot climates. The latest generation of AI processors operates at temperatures near 113 degrees Fahrenheit, which is warm enough to be cooled with water. Parker says this is significantly hotter than traditional systems, reducing the need for water-based cooling. Unless outdoor temperatures regularly exceed 113 degrees Fahrenheit, that threshold is generally reached only during severe heat waves. In most conditions, “large, efficient fans are usually enough to cool the infrastructure and allow the ambient air to move the heat away,” Parker says.
Source: arstechnica.com