Cray and SGI both announced deals that are bringing their computing solutions to institutions conducting research into such areas as weather prediction and earthquakes.
Cray officials on June 26 said that the European Centre for Medium-Range Weather Forecasts (ECMWF) will use two Cray XC30 supercomputers and the company’s Sonexion storage technology in its next operational facility, with the computing system expected to be put into production next year.
The deal, which includes not only the equipment but also multiple years of services, is worth more than $65 million, according to Cray officials.
The XC30—which had been known as “Cascade”—was introduced late last year and is powered by Intel’s Xeon processors, and can also leverage either Intel’s Xeon Phi coprocessors or Tesla GPU accelerators from Nvidia. The coprocessors and accelerators are designed to boost the performance of the supercomputer without significantly driving up the power consumption.
The ECMWF, an intergovernmental organization supported by 34 countries in Europe, develops and operates global models and data-assimilation systems of Earth’s dynamics, thermodynamics and composition. The agency generates a range of global weather forecasts and also conducts research into such areas as wave forecasts and global atmospheric composition.
ECMWF officials said the addition of Cray’s XC30 supercomputer and Sonexion storage will give a boost to the group’s work.
“The Cray system will be a strong foundation to meet our strategic, operational and research objectives,” Alan Thorpe, ECMWF’s director-general, said in a statement. “ECMWF’s history of using supercomputers dates back to 1979 when the Centre issued its first operational medium-range weather forecast.”
The supercomputing infrastructure at the organization will run multiple petaflops of performance, according to Cray officials. It will include two XC30 systems and a multi-petabye Sonexion storage system, which is an integrated, scale-out Lustre system aimed at such workloads as big data and high-performance computing (HPC). It’s designed for high density and can reduce the storage space requirements for petascale systems by as much as 50 percent.
The XC30 supercomputer includes the Aries system interconnect, which is optimized for HPC environments. Cray had been developing Aries until April 2012, when Intel announced it was buying Cray’s networking assets—including Aries and Cray’s Gemini technology—for $140 million.
The supercomputer also comes with Cray’s Dragonfly topology, which helps create low-latency networks, the high-performance Cray Linux Environment, the vendor’s HPC-optimized programming environment, and an air-cooling system in which air is blown into one end of the a row of computers and out the other end, eliminating the traditional hot aisle/cold aisle setup in many data centers.
Cray, SGI Land Deals for HPC Systems
For its part, SGI officials said the deployment of their Rackable servers by Princeton University’s Department of Geosciences is the latest proof-point of the growing demand for GPU accelerators in HPC environments.
The university is installing a range of 2U (3.5-inch) Rackable servers that each include four Nvidia Tesla K20 GPU accelerators, creating a system that will leverage 200 of the accelerators in all. The Department of Geosciences conducts research into earthquakes, and with the new Rackable servers, the time required for reports on seismic activity through the Global Seismicity Portal went from as long as eight hours down to 15 to 30 minutes.
“The need for computing power at lower power consumption continues to be a key customer requirement and customers are benefitting from the highly accelerated code and leading performance per watt capability offered with Nvidia Tesla GPUs,” Bill Mannel, vice president of product management for server products at SGI, said in a statement. “Customers are developing finer meshes, larger models and problems, higher sampling rates and resolutions. These problems are often being solved using Tesla accelerators.”
SGI is seeing a 30 percent year-over-year increase in shipments of Nvidia GPUs, and officials said they plan to grow their GPU capabilities by adding liquid-cooled GPU server blades to its ICE X supercomputers. The company offers Tesla K10, K20 and K20X GPU accelerators in its UV 2000 systems, various Rackable servers and the ICE X.
The deal with Princeton is the latest in a string of customer wins for SGI. Most recently, On June 17, SGI announced that the Department of Defense had deployed an ICE X HPC system for its Spirit supercomputer at the Air Force Research Laboratory, making it the 14th fastest supercomputer in the world, according to the Top500 list. In addition, the vendor on June 4 announced that the Translational Research Institute in Australia is using SGI’s UV 200 shared-memory platform, Rackable compute cluster and InfiniteStorage systems to create a big data HPC solution for medical research at a new facility.