Xeon Phi Supercharging Green HPCs Share your comment!


In a turning point in the high-performance computer (HPC) market, Intel has started delivery of its Xeon Phi many-integrated-core (MIC) processors, set for volume production by Jan. 2013. By incorporating as many as 60-cores into each Xeon Phi coprocessor, future supercomputers multiply performance at a fraction of the current cost and power consumption, solidifying Intel’s lead in green HPCs for the foreseeable future.

Another Boost for Intel’s Skyrocketing HPC Growth

Intel has been steadily increasing its share of the supercomputer market, according to the Top500 Supercomputer List, from a few percent in 2000, when the list was dominated by IBM, to over 75% in 2012.  With the introduction of the massively parallel Xeon Phi coprocessor, housing up to 60 cores per chip, Intel’s share of the HPC market is destined to increase even further. Key advantages include lower cost, lower power and the ease with which x86 parallel processors now be built and programmed using tools like Intel’s Parallel Studio XE 2013 and Cluster Studio XE 2013.

The Xeon Phi aims to be the right solution at the right time for HPC. As the first offering in the MIC architecture, Phi offers green energy efficiency for x86 program execution, thanks to core’s reduced clock speed and the ability to expand parallel processing capabilities at 60-cores per added Xeon Phi coprocessor card. IBM pioneered the technique of reducing core clock speeds to increase energy efficiency for massively parallel supercomputers. Now, Intel is following suit, enabling faster HPCs that consume less power.

The Top500 Supercomputer List reflects both these trends, with seven supercomputers using the Xeon Phi on the list this year, including one in the fastest Top10 (Stampede at the Texas Advanced Computing Center) and another ranked as the most energy efficient (Beacon at the National Institute for Computational Sciences).

For future HPC builders, Intel announced two versions of its Xeon Phi at Supercomputer 2012 (SC12, Nov. 12-15, Salt Lake City), both of which work in concert with a Xeon E5 main processor.

For capacity-bound workloads, such as content creation and energy research, Intel announced its Xeon Phi 5110P co-processor, which offers just over 1-teraFLOPS performance and supports 8-Gbytes of memory with 320 Gbytes per second bandwidth. The Xeon Phi 5110P consumes only 225 Watts and does not require a fan, making it ideal for densely packed HPCs.

New Family Targets Heavy Workloads

The new Xeon Phi 3100 coprocessor family is specifically targeted at workloads that are “compute-bound”, such as life-science applications like protein folding and financial simulations like stock market forecasting. The Xeon Phi 3100 offers 1-teraFLOPS (1000 gigaFLOPS) and support for 6-Gbytes of memory with 240-Gbytes per second bandwidth–all at less than 300 Watts power consumption.

The Xeon Phi 5110P is shipping today to select customers, with general availability slated for Jan. 28, 2013 at $2,649. The Xeon Phi 3100 coprocessor will be available during the first half of 2013 at less than $2,000.

Intel’s Xeon Phi now comes is two flavors–the 3100 (top) which offers screaming speed at 300 Watts, and the the 5110P (top) which at 225 Watts does not require a fan.

Source: Intel

Posted on November 14, 2012 by R. Colin Johnson, Geeknet Contributing Editor