Supercomputer makers Cray and SGI got some good news out of Europe this month.
Cray officials have announced that the Swiss National Supercomputer Centre (CSCS) has installed a Cray CS-Storm, a cluster with a peak performance of 15 teraflops (trillions of lfloating point operations per second) per node that will be used by the Swiss Federal Office of Meteorology and Climatology (MeteoSwiss) to create more detailed weather forecasts.
The system is powered by Xeon server processors from Intel and Nvidia's Tesla K80 GPU accelerators. Cray introduced the CS-Storm systems a year ago, saying at the time that the systems could offer up to eight Tesla K40 GPUs and two Intel E5 processors per each 2U (3.5-inch) computer server, and that each standard 48U (84-inch) rack can hold 22 of the servers. It also is liquid-cooled, which means it won't put more demand on the center's cooling infrastructure.
According to Cray officials, this is the first time a supercomputer powered by GPU accelerators has been used to run production numerical weather models for a national weather service. The Swiss national weather service will use the supercomputer to run advanced simulations and up to eight forecasts every day, more than what it was doing before, officials said.
CSCS Director Thomas Schulthess noted that high-quality weather forecasts rely on a lot of processing power, and with the Cray supercomputer cluster, the center will be able to increase the performance of simulations 40-fold while staying with the same power and space footprint of the system installed for the weather service three years ago. In addition, the GPU accelerators mean that the supercomputer center will spend three times less on power than a system running only on CPUs.
"Today's weather and climate models are ingrained with massive amounts of data and science that have increasingly demanding compute requirements," Barry Bolding, senior vice president and chief strategy officer at Cray, said in a statement. "With an eight-to-two ratio of GPU accelerators to CPUs, the Cray CS-Storm system for MeteoSwiss is a powerful tool for running production models at a much higher resolution and granularity."
Supercomputer makers are gravitating toward accelerators—such as GPUs from Nvidia and Advanced Micro Devices, Intel's x86 Xeon Phi co-processors or field-programmable gate arrays (FPGAs)—alongside traditional CPUs as a way of significantly increasing the performance of the systems while holding down the power consumption. In the latest Top500 list of the world's fastest supercomputers released in June, 88 systems used accelerator technologies, a jump from the 75 on the list released in November 2014. Fifty-two used Nvidia GPUs, four used ATI Radeon graphics from AMD and 33 used Intel's MIC Xeon Phi co-processors.
Cray, with 71 systems, accounted for 14.2 percent of the supercomputers on the list. SGI, with 29, accounted for 5.8 percent.
SGI this month first announced the unveiling of the Salomon supercomputer, a massive system deployed by the IT4Innovations national supercomputing center in the Czech Republic that is based on SGI's ICE X technology. The Salomon, which has a theoretical computer power of 2 petaflops (quadrillion floating point operations per second), is ranked number 40 on the Top500 list of the world's fastest supercomputers and offers more than 20 times the performance of IT4Innovaitons' other system, Anselm, according to SGI officials.
The system is powered by 24,192 Intel Xeon E5-2680 v3 cores with 129TB of memory and 52,704 Xeon Phi 7120P cores with 13.8TB of memory. It also brings 2PB of disk capacity and 3PB of backup tape capacity.
The Salomon supercomputer will run a range of research projects in such fields as cosmology, astronomy, engineering and structural mechanics of liquids, climatology and molecular modeling, according to SGI officials.
SGI also announced that one of its UV supercomputers was used by scientists at the Linding Lab in Denmark in their discovery of how genetic diseases, such as cancer, attack the protein-signaling networks that control human cells, which could lead to improved treatments and medicines. Researchers from the universities of Copenhagen, Yale, Zurich, Rome and Tottori—in Japan—have used the SGI supercomputer to determine how mutations such as those in cancer can damage the protein-signaling network within human cells.
SGI's UV server at the Linding Lab—which is housed at the Biotech Research and Innovation Centre—is powered by Intel's Xeon processors.