Trending Topics

Is liquid cooling imperative for high performance computing?
With AI and compute-intensive workloads demanding more compute performance than ever, liquid cooling could be the technology that saves datacenters from meltdown

Edge-to-cloud, built to transform your business. Learn more about it here.
Heat dissipation is a perennial problem for computers, whether they be the phones in our pockets, the laptops we carry in our bags or the workstations we use at our desks. But that challenge becomes significantly more difficult and serious in the datacenter, where thousands of servers are crunching their way through myriad workloads 24/7.
While traditional air cooling has historically been suitable for the datacenter, it is running out of effectiveness and capacity. The huge increase in power demands – particularly the GPUs in AI accelerators – is pushing this cooling technique to its limits. A typical HPC rack now demands between 20kW and 80kW compared to around 5kW a decade ago.
This has a global impact. Around 2% of the world’s electricity is now powering datacenters, according to estimates, but that’s set to increase to 3% by 2027 and perhaps even hit 5% by 2030, according to research by Goldmans Sachs.
Right now, 40% of a datacenter’s power usage is dedicated to cooling. As our reliance on datacenters increases, it’s clear that more efficient cooling can drastically reduce power consumption, overall operational cost, and usher in ever-more performant compute solutions.
Direct liquid cooling
Liquid cooling provides significant advantages over traditional air cooling in terms of efficiency, power usage, and scalability. That last part is vital, given that even today air-cooling struggles to dissipate the heat generated by the dense number of CPUs and GPUs packed into enterprise compute units. That density is only going to increase.
We need more efficient solutions such as direct liquid cooling. Here, cold plates are positioned directly over the components – much like a heatsink in a PC – then liquid coolant is pumped through the plates drawing heat away from the components. The liquid coolant is then routed through a heat exchanger, where the heat absorbed from the components is transferred to a water system, after which the liquid coolant is routed back towards the components.
100% fanless direct liquid cooling
Direct liquid cooling has been available across a wide range of HPE server hardware for years, with the core components being liquid cooled while ancillary components were air cooled. This hybrid approach typically achieves about 70% of the cooling by liquid and 30% by air.
HPE has been innovating liquid cooling for over 50 years, and it shows no signs of stopping. Its latest innovation is the industry’s first 100% fanless direct liquid cooling solution for servers, essentially removing air cooling from the equation. With every part of the server – CPUs, GPUs, memory, storage, interconnects, network fabric – now liquid cooled, HPE realized significant power efficiency and performance gains. Not to mention reducing power costs and lowering environmental impact.
“As organizations embrace the possibilities created by generative AI, they also must advance sustainability goals, combat escalating power requirements, and lower operational costs,” said Antonio Neri, President and CEO of HPE at the time. “This direct liquid cooling architecture can yield 90% reduction in cooling power consumption as compared to traditional air-cooled systems.”
The world’s top supercomputer: El Capitan

In November 2024, Lawrence Livermore National Laboratory unveiled El Capitan. Built using HPE Cray EX supercomputers, which included 100% fanless direct liquid cooling architecture. El Capitan instantly became the world’s fastest supercomputer, pushing Frontier – also built by HPE – into second place in the TOP500 List of supercomputers. But El Capitan wasn’t just the fastest computer on Earth, it is also in the top 25 Green500 supercomputers.
“El Capitan marks another significant milestone in exascale supercomputing, bringing monumental performance, energy efficiency and the capabilities to accelerate AI-driven scientific discovery,” said Trish Damkroger, Senior Vice President and General Manager, HPC & AI Infrastructure Solutions, HPE.
Enterprise servers
You can bring the benefits of full direct liquid cooling to organizations, too, as it helps to unleash a level of performance that air-cooling cannot match. Any enterprise looking to increase compute performance, reduce operational costs, and improve sustainability should be looking to direct liquid cooling when considering datacenter hardware.
HPE’s latest ProLiant Compute Gen12 servers are available with direct liquid cooling, providing performance, density and infrastructure advantages. Greater performance comes from Intel’s Xeon 6 processors, with up to 288 cores and 8TB of memory, not to mention support for faster PCIe Gen5 storage and connectivity.
There are many other benefits too: quantum resistant encryption, hardware protection thanks to HPE’s “silicon root of trust” and iLO 7 with Secure Enclave, much improved power efficiency and sustainability.
Liquid cooling: No longer a nice-to-have option
Lowering power consumption and heat dissipation can reduce the power to operate, the power to cool and the carbon footprint. As an example, HPE estimates that a modeled datacenter with 10,000 air cooled servers would emit 8,700 tons of CO2 into the environment. If the same facility used direct liquid cooling and upgraded the cooling infrastructure, it would produce just 1,200 tons of greenhouse emissions, a whopping 87% reduction in power and carbon.
And efficiencies can be driven further with liquid cooling. With air, it is difficult to reuse the heat generated. Heat removed in liquids can actually create additional efficiencies. Increasingly, users of liquid-cooling exploit the heat removed in aquaculture, hydroponics, district heating, swimming pools, and industrial process.
Liquid cooling is no longer a “nice to have” option, it’s becoming integral for AI workloads. “If we can’t cool down chips fast enough, datacenters could face overheating issues that will cause system failure and ultimately lead to unplanned downtime on running AI jobs,” said Jason Zeiler, Liquid Cooling Product Manager at HPE. “Liquid cooling is the ideal solution for AI datacenters.”
