Intel, HP Using Xeon Phi Chips for Energy-Efficient Supercomputer
Intel and Hewlett-Packard are working together to build a highly energy-efficient supercomputer that will include the chip maker's upcoming Xeon Phi coprocessors and a design that uses warm water to cool the servers.
The supercomputer, announced Sept. 5, will be used by the Department of Energy's National Renewable Energy Laboratory (NREL) for research into numerous energy-related issues, including renewable energy and energy-efficient technologies, according to Intel officials.
The $10 million system, which will include about 3,200 Xeon processors and another 600 or so Xeon Phi coprocessors and include various ProLiant servers from HP, will eventually offer total peak performance of more than 1 petaflop (a thousand trillion floating-point operations per second) while driving up the energy-efficiency rating of NREL's data center.
Installation will begin in November, with full compute capacity coming online in the summer of 2013, according to the chip maker.
"The heart of NREL is based on a powerful combination of the Intel Xeon processor E5 product family, which leads the data center industry in performance per watt, and Intel Xeon Phi coprocessors which are setting new records for energy efficiency," Raj Hazra, vice president and general manager of Intel's Technical Computing Group, said in a statement. "We are proud that the very best energy-efficient processing technology in computing is the foundation for the supercomputer that will drive the research for renewable energy and energy-efficient technologies."
The Xeon and Xeon Phi technologies are key parts of Intel's high-performance computing (HPC) efforts. The company has targeted HPC as a key growth area-along with cloud and networking-that promise as much as 20 percent annual increases. Intel in March rolled out its Xeon E5-2000 processors, which offer up to eight cores, 80 percent better performance over the previous generation and 50 percent better energy efficiency, with executives targeting the portfolio at HPC and supercomputing.
Xeon Phi is the brand that Intel has wrapped around its Many Integrated Core (MIC) technology, which has been in development for more than two years. Intel officials are fashioning the Xeon Phi chips as co-processors that will work with CPUs such as Xeons to bring parallel-processing capabilities to particular applications, enabling them to run faster than they would on traditional Xeons while consuming less power. The Xeon Phi chips, built on Intel 22-nanometer manufacturing process, will have more than 50 cores when they are released later this year.
Accelerators like the Xeon Phi are becoming increasingly popular in such areas as HPC and supercomputing as more applications are designed to take advantage of their parallel processing capabilities and their energy efficiency. Most of the focus with accelerators has been on graphics chips from vendors like Nvidia and Advanced Micro Devices, but Intel instead is focusing its efforts in the area on the x86 architecture. Intel executives have argued that having x86-based coprocessors offers an advantage over GPUs because of their ability to run more existing code.
"At NREL, we have taken a holistic approach to sustainable computing," Steve Hammond, NREL computational science director, said in a statement. "This new system will allow NREL to increase our computational capabilities while being mindful of energy and water used. We will take advantage of both the bytes of information produced and the BTUs produced."
The supercomputer, which will be housed at the Energy Systems Integration Facility being built in Colorado, will leverage current Xeon E5-2670 processors, future 22nm chips built on Intel's Ivy Bridge architecture and the Xeon Phi chips. It will include ProLiant SL230s and Sl250s servers from Intel powered by the current Xeon E5-2670s, and next-generation systems that will run on the upcoming Ivy Bridge and Xeon Phi chips.
Between the computing technologies from Intel and HP and the warm-water cooling capabilities, the new data center will be among the most energy-efficient, according to Intel and NREL. Applying the power usage effectiveness (PUE) standard, which measures the energy efficiency of a building, the data center should get a PUE rating of 1.06. The ideal PUE rating is 1.0; an average data center PUE rating is about 1.92, according to the Environmental Protection Agency.
The cooling will be done by sending warm water into computing racks to absorb the heat coming from the system. The water, which will run as high as 95 degrees, will then be circulated to heat an office and lab space next door, or to heat other parts of the NREL campus. The design for the cooling system was created by people from Intel, HP and NREL.
System vendors have been using water and other coolants for several years to help remove heat from data center facilities. Water cooling tends to be more efficient than air. Intel this week announced it is conducting tests to determine if PCs can be submerged in a mineral oil solution from Green Revolution Cooling.