Intel demos one teraFLOPs performance over Knights Corner
Chip maker steams ahead in quest for Exascale computing
Intel has demonstrated a single Knights Corner coprocessor delivering over one TeraFLOPs of performance, as measured by the industry standard benchmark Linpack, at the International Supercomputing Conference in Hamburg.
Knights Corner coprocessors, which will not start shipping until the second half of 2012, contain more than 50 cores and are based on Intel’s 22nm process technology. They will be available in a PCIe form factor offering 8GB of GDDR5 memory, and will support Intel’s standard programming models.
Raj Hazra, Intel Corporation VP and general manager of the Technical Computing at Data Center and Connected Systems Group, said that Knights Corner coprocessors will be capable of running complete applications as a virtually networked node in a cluster.
Related Articles on Techworld
“We see this capability not only providing performance but significantly easing the porting and tuning of applications today that run on standard clusters, but could benefit from a highly parallel architecture,” he said.
The first Knights Corner-based development cluster is already up and running, and ranked 149th on the Top500 list, delivering 118 TFLOPs of performance. Each node consists of one Knights Corner connected to two E5 processors.
When Knights Corner launches later this year it will become part of the newly-named Xeon Phi product family, based on its Many Integrated Core (MIC) architecture.
Hazra said that Xeon Phi will complement Intel’s existing Xeon E5-2600/4600 product line. While the first generation primarily targets high performance computing (HPC), future generations of Intel Xeon Phi products will also address enterprise data centres and workstations.
“The Phi portion of the brand adds an exciting element to what MIC brings to the overall Xeon family,” said Hazra. “It is optimised to deliver the highest levels of performance from an architecture that is essentially the workhorse for highly parallel compute in the future.”
Hazra added that the introduction of the Phi brand was an important step in Intel’s move towards Exascale computing. The company predicts that the growth of HPC processor shipments will reach more than 20 percent CAGR in the next 5 years.
“What’s driving this growth is HPC has become an essential equity for business and in fact national competitiveness as well. In essence, to compete you must compute,” he said. “It took us four years to go from one petaFLOPs to 10 petaFLOPs, and we’ll see shortly that is will go from 10 to 20 petaFLOPs in one year.”
The Intel Xeon Phi coprocessor is supported by 44 manufacturers including Bull, Cray, Dell, HP, IBM, Inspur and NEC. Cray today announced that its next-generation supercomputer code-named ‘Cascade’ will run on Intel Xeon Phi coprocessors.