Frugal systems key to data centres of future
Data centres powering artificial intelligence face a major sustainability challenge in preserving water, according to Prof Jeff Punch of the University of Limerick engineering department.
A recent UN report warns that AI data-centre water usage, cooling massive, heat-generating computing workloads, could match the daily needs of 1.3 billion people by 2030, he pointed out.
Data centres now draw such vast quantities of electrical energy, almost a quarter of metered consumption, that it is imperative to reduce consumption and reuse the resource where possible.
Evaporative cooling towers
Traditional air-cooled data centres rely on evaporative cooling towers, using water's high latent heat to extract heat from server rooms, the prof writes.
While this saves electricity, it consumes vast amounts of water.
Water excels at thermal regulation in the same way that sweat cools human skin.
As water evaporates in cooling towers, it extracts heat efficiently.
However, this process consumes immense amounts of fresh water, turning a cooling solution into a resource strain, the professor explains.
To reduce water reliance while handling rising processing demands, alternative technologies offer higher efficiency and lower resource consumption.
These include direct-to-chip liquid cooling, closed-loop dry cooling, immersion cooling, and waste heat re-use.
Retrofitting costs
Transitioning away from conventional air and evaporative cooling carries upfront retrofitting costs and requires retention of air cooling for older equipment.
However, high energy draw makes adopting frugal, liquid-based cooling systems imperative for future data infrastructure, the academic points out.