Technical analysis documents accelerating adoption of liquid cooling systems for power-dense datacenter GPU operations.

Traditionally reserved for mainframes and academic supercomputers, liquid cooling is beginning to move into enterprise data centers as more demanding workloads push up power densities, leaving operators seeking more efficient alternatives to air-based cooling systems. Hyperscale cloud operators, including Microsoft, Alphabet's Google, Facebook, and Baidu, have formed a group working on an open specification for liquid-cooled server racks. Google has publicly stated that its TPUs are now cooled using a direct-to-chip liquid cooling design. Despite operators' reluctance to disclose specific applications—viewing them as competitive advantages—usage trends are starting to emerge across several categories of workloads.

Hardware accelerators, particularly GPUs alongside high-performance CPUs, represent a primary driver of liquid cooling adoption. GPU-powered machine learning is the most common use of hardware acceleration outside of the HPC realm, though about a third of IT service providers surveyed by 451 Research reported plans to accelerate systems for online data mining, analytics, engineering simulation, video, live media, fraud detection, load balancing, and latency-sensitive services. These accelerators have thermal design points much higher than CPUs, typically requiring 200W or more of cooling, with individual systems needing more than 1kW of cooling. Intel is also pushing beyond the 150W limit it traditionally designed server processors to. "More and more people want more powerful chips, and we're starting to see the watts creeping up," according to Andy Lawrence, an executive director at the Uptime Institute. Most data centers tracked by Uptime now have at least some racks exceeding 10kW, with 20 percent having racks at 30kW or higher. As Lawrence explained, "If people are putting a GPU in alongside an Intel processor, they'll probably get three times the density they were previously getting."

Storage and edge data centers represent emerging adoption areas for liquid cooling. Newer storage technologies, including Solid State Drives and helium-filled HDDs, can be cooled with full-immersion solutions, allowing dense storage to be co-located with liquid-cooled processing. The combination of SSDs and helium-filled HDDs also eliminates the need to separate air-cooled storage from liquid-cooled processing, while offering improved HDD reliability by reducing heat and humidity effects. Edge data centers—densely populated remote facilities at wireless towers, factory floors, and retail stores—are increasingly hosting high-density compute hardware such as GPU-packed clusters for machine learning. In Lawrence's estimate, as much as 20 percent of edge data centers could use liquid cooling, with his vision encompassing lights-out micro-modular high-density sites supporting 40kW per rack.

Financial services workloads also drive liquid cooling demand, particularly computationally intensive high-frequency trading systems and blockchain-based applications like smart contracts and cryptocurrencies. GRC, formerly called Green Revolution Cooling, has a high-frequency trading firm testing its immersion-cooling solution. The vendor experienced its biggest spike in sales when it introduced an immersion-cooling product for cryptocurrency mining coinciding with bitcoin's price surge beginning in late 2017. A GRC customer in Trinidad and Tobago is running a cryptocurrency service at 100kW per rack with a warm water-based cooling loop connected to an evaporation tower, according to GRC CEO Peter Poulin. The use of warm-water cooling over cold-water cooling improves energy efficiency, enabling the service to operate in tropical climates.