The U.S. government has awarded IBM a contract to build a supercomputer capable of performing at 20 petaflops, which is more powerful than all of the systems on today's Top500 supercomputer list combined. Terms of the deal were not disclosed.
Nuclear researchers at the U.S. Department of Energy will use the planned Sequoia system, to be built in IBM's plant in Rochester, Minn.
The fastest systems today perform at a little over 1 petaflop, or 1 thousand trillion floating-point operations per second. The single-petaflop barrier was passed last May by IBM's Roadrunner supercomputer.
Sequoia will be housed at the Lawrence Livermore National Laboratory. Mark Seager, assistant department head for advanced technology at Lawrence Livermore, called the new system “the biggest leap of computing capability ever delivered to the lab.”
The Linux-based Sequoia system will use approximately 1.6 million IBM Power processors. IBM is still developing a 45-nanometer chip for the system and may produce a processor with eight, 16 or more cores. The system will have 1.6TB of memory and will be housed in 96 refrigerator-size racks.
The Sequoia contract also calls for IBM to deliver a smaller, 500-teraflop system, called Dawn, to the Livermore, Calif.-based lab. Dawn will be operational later this year, and researchers will use it to prepare for the larger system.
The national lab's work with Sequoia will likely pave the way for broader adoption of massive systems that could improve weather research, forecasts, tornado tracking and work on a variety of other research problems, officials said.
