Cray to build supercomputer for US nuclear scientists

Cray is building one of the world's fastest supercomputers at Los Alamos National Lab, in a $45 million government contract. The supercomputer will be a petascale machine, meaning it will reach speeds of at least one thousand trillion calculations per second, or a "petaflop." Only two computers operating today reach such speeds, according to the latest Top 500 Supercomputer Sites list, released in November.

The world's fastest machine is a Cray XT5 supercomputer known as Jaguar, which is at the Department of Energy's Oak Ridge Leadership Computing Facility and uses 250,000 processing cores to reach speeds of 1.75 petaflops. The newly announced Los Alamos computer, named Cielo, uses a new interconnect chipset and software from Cray that "improves the performance, productivity, and reliability" of the system architecture found in XT5.

Cray is not saying exactly where Cielo will rank among the world's top supercomputers when it goes online the second half of 2010, promising only that it will top the petascale barrier, a milestone first reached in June 2008. 

Los Alamos National Laboratory, established in 1943, is one of the most prominent consumers of supercomputing technology, using the huge machines to safeguard the nation's nuclear weapons stockpile and perform other scientific tasks. Los Alamos hosts the second fastest machine in the world today, an IBM cluster known as Roadrunner that operates at petaflop speed. IBM and Los Alamos are trying to smash all records with a 20-petaflop machine, but that system won't be ready until 2012.

In addition to Los Alamos, Cielo will be used by two other National Nuclear Security Administration labs, the Sandia National Laboratories and Lawrence Livermore National Laboratory. The machine will be housed at Los Alamos's Strategic Computing Complex.

"The NNSA will use the new supercomputing system to ensure the safety, security and effectiveness of the United States' nuclear stockpile, and will run the NNSA's largest and most demanding modeling and simulation workload," Cray says.