Dell, eyeing the potential of liquid cooling for servers, has developed a technology to detect leakage of liquid cooling modules, according to Eric Leung, director of Enterprise Solution at Dell EMC.
Since traditional air cooling methods can only achieve optimized power usage effectiveness (PUE) of 1.4, liquid cooling has become necessary if server companies wish to bring the number down to below 1.2, Leung said.
Liquid cooling currently still accounts for less than 1% of overall cooling solutions adopted by servers, but demand for such solutions has been picking up as performances of server CPUs and GPUs have been rising, prompting brand vendors to adopt more effective cooling methods for their servers.
Intel's upcoming 10nm Ice Lake server processors are able to compute data 1.5-8 times faster than the previous-generation CPUs, but their power consumption will also rise from 150-250W of the previous-generation products to 250-300W.
AMD's next-generation Milan-based EPYC 7763 server processor combined with its Instinct MI100 HPC GPU accelerator reportedly will also consume up to 280W of power.
Nvidia's Ampere-based A100 GPU features a performance boost of 20 times compared to previous-generation Volta-based GPUs, targeting primarily clients' demand for AI applications.
Alibaba has already adopted submersion-type liquid cooling solutions for its server products, while Microsoft, Google and Facebook are also considering adopting liquid cooling for their datacenters.
Open Compute Project (OCP), led by Facebook, reportedly will use the liquid cooling back plate solution as a standard cooling method. Microsoft and Google are planning to adopt two-phase immersion cooling modules for their servers.
