A University of Texas of supercomputer will tap a future Intel chip that contains more than 50 processor cores--the first instance of Intel supplying this novel technology to a commercial computer.
The Texas Advanced Computing Center (TACC) and University of Texas announced today that they will deploy a 10-Petaflop (or 10,000 trillion operations per second) supercomputer dubbed "Stampede."
When it arrives in early 2013, the supercomputer is expected to be among the world's most powerful computers for scientific and financial applications.
Inside will be an Intel chip design codenamed "Knights Corner," which will house more than 50 processor cores. Intel server chips typically top out at about 8 cores.
But Knights Corner is not a typical Intel chip. It has its roots in a project called Larrabee, which initially was intended to be a many-core, high-end graphics processor unit or GPU for gaming and media applications. That project was canceled in 2009.
GPU cores are typically smaller and more specialized than CPU (central processor unit) cores--the latter is what Intel has traditionally manufactured for the PC market. GPUs can be much faster than CPUs at specialized tasks. (Though even the definition of an Intel CPU is changing. Intel's Sandy Bridge CPU, for example, integrates a many-core GPU.)
Intel has gotten serious about many-core processors because Nvidia--traditionally a supplier of GPUs--has become a force in the supercomputer market with its Tesla processors. Nvidia stunned observers in 2010 that was one of the fastest at the time.
Intel describes Knights Corner as "a co-processor aimed at highly parallel workloads...and will be built using Intel's leading edge 22 [nanometer] 3D Tri-Gate transistor technology when in production."
"My group...is excited about the opportunities Stampede offers to greatly accelerate our work in quantifying uncertainties in computer models of dynamics of polar ice sheets, global seismic wave propagation, and whole-earth plate tectonics," said Omar Ghattas, director of the Center for Computational Geosciences in the Institute for Computational Engineering and Sciences at the University of Texas at Austin, in a statement.
When completed, Stampede will be composed of several thousand Dell "Zeus" servers with each server having dual 8-core processors from the forthcoming Intel Xeon processor E5 Family (formerly code-named "Sandy Bridge-EP") and each server will have 32 gigabytes of memory. This production system will offer almost 2 petaflops of peak performance. The cluster will also include the Knights Corner chips providing an additional 8 petaflops of performance, the Texas Advanced Computing Center said in a statement.