Global demand for technologies such as AI, high-performance computing (HPC) and cloud services is accelerating, and the efficiency of the cooling systems used in the data centers behind such technologies is becoming increasingly important.
Today, data centers account for around 1.5% of global electricity use: by 2030, technologies such as AI will mean energy demand will double worldwide, according to the International Energy Agency.
Director for EMEA HPC/AI at Lenovo.
The conventional approach to cooling data centers is air cooling, with large fans inside the data centers and inside the devices. As power demands increase, air cooling is struggling to meet the demands of current and future data centers.
Liquid cooling, which uses hot water to cool components, is becoming an important alternative, with key advantages in terms of efficiency and sustainability.
With AI workloads demanding ever-increasing power densities and sustainability moving to the forefront of infrastructure strategy, it is key for data center leaders to make informed cooling decisions. The following are the main advantages and disadvantages of the two approaches, considered based on multiple factors.
Thermal efficiency
One of the key advantages of liquid cooling systems is efficiency, where water can provide air with much higher efficiency. Water is more than 3,000 times more efficient at removing heat than air, meaning data center operators have to use less electricity to keep equipment and data centers cool.
Liquid cooling techniques, such as direct-to-node cooling, can remove up to 98% of heat from servers, and hot water can also be reused to heat facilities such as buildings and even swimming pools. Overall, liquid cooling can reduce power consumption by up to 40%.
Sustainability
Liquid cooling and air cooling have advantages and disadvantages in terms of sustainability, although liquid cooling technology is evolving rapidly. Air cooling consumes a lot of energy, but saves water.
Liquid cooling is more energy efficient, but older systems that run on evaporative cooling, where hot water is sprayed on the pads to cool them, must be replenished from external sources.
Hot water cooling systems reduce water loss and newer systems can accept high inlet temperatures, meaning less energy is wasted cooling the water.
Organizations are also moving from open-loop systems for heat management (where water is cooled through evaporation) to closed-loop systems (which remove heat through a closed-loop liquid-air heat exchanger) to make liquid cooling more efficient and scalable.
Spatial density
Air-cooled systems require greater physical space to operate, and air-cooled systems can support up to approximately 70 kilowatts per rack, but there is a physical limit known as “specific heat capacity” beyond which they cannot operate.
The GPUs that power AI platforms require up to 10 times more power than traditional CPUs and contain a higher number of transistors: other design features, such as 3D silicon stacking, are allowing GPU manufacturers to pack more components into smaller spaces.
This increases the power density of data centers, which is pushing air cooling beyond the threshold where it is viable. As data centers consume more and more power, air limits scalability, meaning less computing power can be placed in the same amount of space.
Liquid cooling allows components to run faster, meaning data centers can run more intense AI and HPC workloads, enabling greater compute density.
Future preparation
According to Goldman Sachs, by the end of this decade, the energy demands of technologies like AI will have driven a 160% increase in data center energy demand. This means that the increased efficiency of liquid cooling will become increasingly important.
Measured in power usage effectiveness (PUE), data center efficiency is the energy needed to run the entire data center, divided by the power requirements of the IT equipment.
Some liquid-optimized data centers are already achieving PUEs of 1.1 or even 1.04, and customers will also be able to use hot water to heat office buildings, which is becoming even easier with the latest systems. As computing demand grows around the world, such efficiencies will become increasingly important.
Reliability and maintenance
Air-cooled systems have clear disadvantages when it comes to reliability and maintenance: the fans used can expose hardware to dust, and temperatures are more likely to fluctuate in air-cooled data centers, which has implications in terms of maintenance.
Generally speaking, liquid cooling systems require more expertise to maintain. But that is changing. Modern liquid cooling systems are useful, safe and very practical in a wide variety of environments.
The latest water-cooled systems can operate in multiple configurations for ease of adoption, including hybrid systems that use air and water.
We have presented the best eco-friendly web hosting.
This article was produced as part of TechRadarPro's Expert Insights channel, where we feature the best and brightest minds in today's tech industry. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing, find out more here:
