After spending billions to find out it won't be able to compete with AMD or Nvidia

Apr 9, 2012 09:45 GMT  ·  By

World’s biggest CPU manufacturer, Intel, has its people making waves and raising expectations in anticipation of the arrival of the “Stampede” supercomputer next year. Intel’s MIC (Many Intel Cores) is the company’s attempt at building a highly parallel architecture that could, someday, compete with AMD and Nvidia’s GPUs.

In a statement cited by xbitlabs, Intel’s Radoslaw Walczyk said that, “Knights Corner is in great shape and is exactly where it has to be according to our internal schedule. We have not disclosed any information related to production or launch date of 'Knights Corner'.”

Other company representatives, like Diane Bryant, vice president and general manager of Intel's server product group, said that Intel’s MIC was “set to go into production in about a year.”

They are most likely referring to Texas Advanced Computing Center’s “Stampede” supercomputer.

Using funds from a National Science Foundation grant, the Texas Advanced Computing Center, or TACC, from the University of Texas in Austin, announced its plans to build a supercomputer that would be operational in January of the next year.

It will be made of several thousand DELL “Zeus” servers using dual socket platforms with 8-core Xeon E5 CPUs and 32 GB of memory.

The cluster will be using Intel’s MIC co-processors and will provide 10 petaflops of performance.

In the DGEMM benchmark, using a double-precision general matrix-matrix multiplication test, Intel’s Knight’s Corner, with its 50 cores built on 22nm technology, is able to deliver 1 teraflop of performance.

This is almost twice the performance offered by Nvidia’s Tesla 2090 card of around 665 gigaflops. Sure, the Tesla is a one-year-old card and the new Tesla generation is right around the corner, but Intel has quite an achievement here.

Even more impressive is the fact that all those thousands of Xeon E5 CPUs will offer just 2 petaflops of performance, while the “Knights Corner” co-processing units will be responsible for 8 petaflops of computing power.