Cooling Today’s Hot New Processors
This article originally appeared on insidehpc.com
Every now and then the HPC industry undergoes a unique transformationand heads off in a new direction. That’s what is now underway with the introduction of such ground breaking new technology as the Intel Xeon Phi family of powerful, manycore, highly parallel processors.
Expected later in 2016, Intel will be releasing production versions of its Knights Landing (KNL) 72-core coprocessor manufactured on a 14mm process with 3D Tri-Gate transistors. The chip is designed for highly parallel processing and provides double precision performance exceeding 3 teraflops. It also features on-package MCDRAM memory.
Intel’s KNL and upcoming GPUs from NVIDIA are having a direct impact on the design of current and future supercomputers and are being adopted by many of the major OEMs. Knights Landing is also driving extensive code modification of existing and future applications to take advantage of the chip’s unprecedented parallelism.
These next generation processors, GPUs and coprocessors are impacting the physical design of the supercomputers now coming down the pike in a number of ways. One of the most dramatic changes is the significant increase in cooling requirements – these are high wattage chips that run very hot and present some interesting engineering challenges for systems designers.
Asetek Meeting Today’s and Tomorrow’s Challenges
Asetek is designing advanced liquid cooling systems specifically tailored to cool supercomputer systems incorporating these next generation chips, regardless of whether they are CPUs or GPUs. Larry Vertal, Data Center Marketing at Asetek, highlights some of the key consideration when it comes to cooling these high-powered, many core, next generation systems.
Says Vertal, “The supercomputer OEMs all have different board configurations and use a variety of layouts to incorporate the new KNL chips into their system. Because Asetek’s cooling solution is made up of discreet coolers linked together using flexible tubing, we are able to adapt to a wide variety of board layouts and cool components in any order that is required. This has proven to be a highly efficientapproach.”
He adds that other vendor’s approaches to liquid cooling employ the use of metal heat pipes and metal tubing. But, he points out, bending metal is more complicated and can prove to be a stumbling block when it comes to finding a solution in a timely and cost effective manner.
Cooling the Supporting Cast
“Another major consideration is that these high wattage chips are dependent on other components on the board that support the core computing unit,” Vertal continues. “These components themselves are now running a lot hotter and have their own individual cooling requirements. Voltage regulators (VRs) are a good example.”
Design considerations include how to best cool these outboard components as well as the outside of the processing unit itself. Asetek’s flexible approach allows designers to “daisy chain” the components into unified system handled by a single cooling solution. This approach also works equally well for the new, high-powered GPUs.
“Another benefit of taking the Asetek approach to designing its liquid cooling solutions is that it allows us to scale our systems as Intel, NVIDIA and others evolve subsequent generations of chips,” Vertal says. “We can adapt to and service the higher wattage components that were not a part of initial design specifications. Our strategy provides the headroom that allows the OEM to design using one liquid cooling solution – a solution that can scale to meet the demands of higher wattages and heat output.”
Systems like Asetek’s RackCDU D2C™ are a good example of this flexibility in action. RackCDU D2C is a hot water, direct-to-chip data center liquid-cooling system that enables cooling energy savings of up to 80% and density increases of 2.5x when compared to modern air-cooled data centers.
“The evolution of high powered chips is an on-going process,” Vertal concludes. “We have the liquid cooling solutions in place now that can help support this new supercomputing paradigm as it unfolds.”