Department of Energy Chooses Xeon Phi Share your comment!


The Department of Energy’s (DoE) Environmental Molecular Sciences Laboratory (EMSL) has chosen the Intel many-integrated core (MIC) architecture with Xeon Phi coprocessors for its next-generation parallel processor at its Pacific Northwest National Laboratory (PNNL).

The new supercomputer uses Intel Xeon and Xeon Phi cores, and will aim for peak performance of 3.4 petaFLOPS, which is more than 20 times faster than its current four-year-old supercomputer based on AMD cores.

The $17 million cluster supercomputer–whose contract was awarded by the DoE’s Office of Science–is being designed by Atipa Technologies (Lawrence, Kansas) and Super Micro Computer Inc. (San Jose, Calif.) to be one of the Top20 fastest supercomputers in the world

Atipa provides end-to-end solutions using Intel Xeon processors and Xeon Phi coprocessors (center) to extract extreme performance from highly parallel applications. SOURCE: Atipa

The Environmental Molecular Sciences Laboratory at PNNL has been assigned the task by DoE of creating climate and biological simulations that are accurate down to the molecular scale.

The aim is to assist environmental scientists with modeling the chemical processes affecting our climate, to investigate the feasibility of using biology-based fuels to replace fossil fuels, to discover new materials for energy storage/generation, and to perform a variety of other scientific tasks that are computationally intensive.

The supercomputer, which houses 2.7 petabytes of memory, will be able to run gigantic molecular-level simulations.

“Our users conduct experiments which they want to verify with models which are accurate at the molecular level,” said PNNL computational scientist Bill Shelton, the associate director at EMSL who manages high performance computing. “Integrating computational theory with experiment is critical to accelerating scientific discovery.”

Atipa will add acceleration capabilities specifically optimized for large-scale molecular-level simulations, including 195,840 Xeon and Xeon Phi cores on 1440 compute nodes with 128 Gbytes of main memory per node connected by Infiniband.

Each node supports dual octal Xeon processors along with dual Xeon Phi coprocessors running up to 120 extra parallel computations per clock cycle.

All total, the system houses over 23,000 Intel processors with over 184,000 gigabytes of memory–about four times as much memory per processor as other supercomputers, according to Atipa.

The supercomputer will be housed in 42 racks, and will include a file system that offers over 2.7 petaBytes of hard disk memory with a 60 Gbyte-per-second read/write speed, permitting gigantic simulations as required by next-generation biology, climate research, chemistry and materials science simulations.

Posted on by R. Colin Johnson, Intel Contributing Editor