Cray is rolling out the latest generation of its XC supercomputers, which are powered by Intel’s newest Xeon E5 processors and offer twice the performance over the previous systems.
The XC40 also comes with Cray’s new DataWarp I/O acceleration technology, which connects solid-state drives (SSDs) directly to the compute nodes in the supercomputer, reducing the distance data has to travel, driving up performance and improving total cost of ownership, according to Barry Bolding, vice president of marketing and business development at Cray.
The XC40 comes two years after Cray unveiled the XC30, which marked Cray’s move from Advanced Micro Devices’ Opteron chips to Intel’s Xeon processors. Intel launched the new Xeon E5-2600 v3 “Grantley” processors Sept. 8, on the eve of the Intel Developers Forum. The 22-nanometer chips, based on the “Haswell” architecture, come with as many as 18 cores and are armed with DDR4 memory, which offers power and energy efficiency over DDR3.
In addition, the XC40 supercomputer and Cray’s new CS400 cluster system, also introduced Sept. 30, can leverage Nvidia’s GPU Tesla accelerators and Intel’s Xeon Phi co-processors to help drive up performance while keeping power consumption in check.
The liquid-cooled XC40 offers up to 384 sockets and up to 226 teraflops of performance per cabinet. A slightly smaller and less dense XC40-AC is air-cooled and is less costly than the larger system.
The DataWarp technology comes as an option with the XC40, according to Bolding. The technology can scale from 70,000 to 40,000 IOPS (input/output operations per second), delivers up to five times the performance of spinning disks at the same cost and helps to handle “bursty” I/O-intensive applications, saving users storage costs. The SSDs are used as cache, temporarily storing data before it’s moved onto the compute or external storage, and are connected to the compute nodes by Cray’s Aries interconnect technology.
It would prove too costly and unwieldy to get the same performance from disks in external storage environments alone, Bolding said. With DataWarp, the organization can create flexible pools of SSD storage that can be leveraged as needed.
“It doesn’t replace the need for spinning disks,” he said in an interview with eWEEK. “Our goal is not to replace spinning disks with solid-state [drives], but to have the right balance.”
The CS400 cluster supercomputer also offers twice the performance of its predecessor and leverages the Nvidia GPU accelerators and Intel co-processors, and offers liquid or air cooling. In addition, organizations can use blade or rack servers in the cluster, which is managed by Cray’s ACE software.
The XC40 system will be the foundation of the Trinity supercomputer planned for the Los Alamos National Laboratory. Cray in July announced the $174 million contract with the National Nuclear Security Administration, with the Trinity supercomputer being used to run massive simulations that will look at the security and effectiveness of the country’s nuclear stockpile. The bulk of the new system will be delivered late next year and into 2016.
Other computing centers that will use the XC40 and CS400 systems include the Korea Meteorological Administration, the National Nuclear Security Administration, the PDC Center for High Performance Computing at the KTH Royal Institute of Technology in Sweden, and the Department of Defense High Performance Computing Modernization Program, according to Cray.