Product and service reviews are conducted independently by our editorial team, but we sometimes make money when you click on links. Learn more.

Liquid Server Cooling Goes Mainstream With Fujitsu PRIMERGY

By - Source: Fujitsu

Data center administrators covet density (in all its forms) above most other considerations because it is one of the key factors to reducing OPEX. Packing more compute into a smaller space either shrinks the overall footprint or provides more processing power within the same amount of space. The problem with incredibly dense architectures is the sheer amount of power piped into each rack. Servers emit power as heat, and dense designs present tremendous cooling challenges.

We have all heard of the cutting-edge open-air hyperscale deployments that provide staggering PUE (Power Usage Effectiveness) metrics, but in the real world, there are very few operators with the scale, engineering talent, and custom architectures to deploy an open-air facility. Facebook, Google and Microsoft (et al), comprise an incredibly small portion of the overall picture, so traditional applications require mainstream solutions.

Fujitsu, which recently announced its DX8700 and DX8900 products, is offering its take on datacenter efficiency by extending its Cool-Central Liquid Cooling Solution to the company's scale-out and HPC offerings. This system utilizes liquid cooling to cool the internal components, thus increasing the cooling efficiency in each rack, which allows for denser designs. Liquid cooling may sound radical and new to some, but it is actually a proven technology in the data center; it was just previously limited to custom HPC/HFT applications.

Fujitsu and Asetek partnered up to deliver the solution, which cools CPUs, GPUs and DIMMs. The system can utilize the heated water (60C/140F) to cool buildings, or merely use ambient outside air to re-cool the liquid without the need for a chiller. According to Fujitsu, the system also reduces power consumption by as much as 50 percent, which will help defray the inevitable higher upfront acquisition cost.

The system is comprised of an OEM'd Fujitsu version of Asetek's RackCDU D2C cooling technology. The example system (pictured at right) utilizes two CPU coolers, and the liquid outputs terminate at the rear of the blade. This allows the system to offer the same serviceability characteristics as a normal server. The example shows the blade slots in the parent chassis and the requisite water connections at the bottom rear of each blade slot. These systems typically use special fittings that retain all liquid when they are disconnected. 

The Fujitsu PRIMERGY cooling system is designed to deliver up to five times more density than a standard deployment, which enables up to 30 kW of power per rack. The higher power ceiling allows for up to 160 Intel Xeon processors and 1,280 memory modules per rack. Fujitsu predicted that by 2016, high-density servers can reach up to 50 kW per rack, and its forward-looking modular design sets the groundwork for handling the increased demands of the future.

Fujitsu designed the system to reduce the PUE ratio to 1.06, which is very close to that attained by the most efficient systems in hyperscale datacenters. Fujitsu introduced the system at the ISC High Performance HPC Event, and it will be available globally in September 2015.