Intel Reveals New Details Of Exascale Supercomputing Chip
Researchers at Intel are building new chip architectures to bring high-performance computing into the exascale range, or 1,000 times faster than a petaflop, which is one quadrillion, or a thousand trillion, calculations per second.
The No. 1 chipmaker disclosed new technical details about its new line of processors, which are based on Intel’s “Many Integrated Core Architecture,” saying its latest technology for high-performance computing is worthy of an entirely new brand – Xeon Phi.
“The first generation of Intel Xeon Phi product family (coprocessors codenamed “Knights Corner”) will complement the existing Intel Xeon processor E5-2600/4600 product families and deliver new levels of performance for highly parallel workloads,” the company announced on Monday during the International Supercomputing Conference (ISC12) in Hamburg, Germany.
“While the first generation primarily targets high performance computing (HPC), future generations of Intel Xeon Phi products will also address enterprise datacenters and workstations.”
Intel’s first generation Xeon microprocessors are currently the most popular calculating engines for conventional servers and supercomputers.
“As we add Intel Xeon Phi products to our portfolio, scientists, engineers and IT professionals will experience breakthrough levels of performance to effectively address challenges ranging from climate change to risk management,” said Raj Hazra, Intel Corporation VP and general manager of the Technical Computing at Data Center and Connected Systems Group.
“This is the next step of Intel’s commitment to achieve exascale-level computation by 2018, and create a unique technology category that delivers unprecedented performance for today’s highly parallel applications.”
The new Xeon Phi processors are expected to be available by the end of the year.
In addition to their compatibility with x86 programming models, the chips will be visible to applications as an HPC-optimized, highly-parallel, separate compute node that runs its own Linux-based operating system independent of the host OS, Intel said. In other words, the chips will work to speed up the fastest clusters by working as a “co-processor” in conjunction with a server CPU to accelerate workloads.
The Xeon Phi uses Intel’s 22nm, 3-D tri-gate transistors, and contains more than 50 cores and a minimum of 8GB of GDDR5 memory, Intel said.
Last year Intel demonstrated the single Knights Corner coprocessor delivering over 1 TeraFLOPs (1 trillion floating point operations per second) of double precision real life performance. By comparison, in 1997, it took more than 9000 Intel Pentium processors inside the ASCII RED supercomputer to break the 1 TeraFLOPs milestone.
Intel’s newest chip has received strong industry support, with 44 manufacturers including Bull, Cray, Dell, HP, IBM, Inspur, SGI and NEC committed to including it in their system roadmaps.
“We are very excited to announce that our next-generation supercomputer code-named ‘Cascade’ will be available with the Intel Xeon Phi coprocessors, giving Cray customers the ability to push the limits of research and discovery,” said Peg Williams, Cray’s senior vice president of high performance computing systems.
“Our Cascade system will feature some of the most advanced and highly innovative HPC technologies ever put into a Cray supercomputer. Combining these features with industry-leading Intel Xeon processors and the new Intel Xeon Phi coprocessors will result in a very compelling system for HPC centers around the world.”
Due to power-on early next year, the first Petascale class supercomputer powered by a combination of the Intel Xeon processor E5 family and Intel Xeon Phi coprocessors will be called “Stampede.”
Intel said it expects a growing number of Petascale-class machines in the coming year that utilize the Xeon Phi coprocessors.
“Reaching Exascale levels of performance requires innovation across the architecture and fabric within high performance computing. To deliver on its promise to reach Exascale performance by 2018, Intel made significant investments in areas that will accelerate its capabilities to deliver this performance in the future,” Intel said.
“The recent acquisition of Infiniband and interconnect assets from QLogic and Cray present key areas for Intel to innovate as the company eliminates the bottlenecks in delivering scalable Exascale-class platforms in the future.”
Intel’s announcement coincides with Monday’s news that IBM’s Sequoia BlueGene/Q System had managed a sustained 16.32 petaflops, or 16.32 quadrillion calculations per second. The accomplishment returns the title of ‘World’s Fastest Supercomputer’ to the United States, who had relinquished the claim two years ago to China.