IBM Building Next Generation of BlueGene Supercomputers

IBM and the Department of Energy's Lawrence Livermore National Laboratory have signed a new contract to build the next generation of IBM's BlueGene supercomputers at the famed DOE facility. The first IBM BlueGene supercomputer, called "Dawn," will have a top processing speed of 500 teraflops. The second IBM system, dubbed "Sequoia," will offer 20 petaflops of performance and surpass the records Big Blue set when it installed the massive Roadrunner system for the DOE in 2008.

IBM and the U.S. Department of Energy have signed a new agreement that will allow Big Blue to build its next generation of BlueGene supercomputers at the Lawrence Livermore National Laboratory in Livermore, Calif.

IBM and the DOE announced their latest partnership Feb. 3. The first part of this new supercomputer installation will begin later this year, and IBM plans to complete the project in 2012, when the new BlueGene system, called "Sequoia," is finished.

When it's complete, IBM is hoping that the Sequoia supercomputer erases the performance benchmarks the company set with its Roadrunner system, which went online in 2008.

The $100 million Roadrunner system, which is currently installed at the DOE's Los Alamos Laboratory in New Mexico, offers a peak performance of 1.105 petaflops, or more than 1.1 quadrillion calculations per second. When Sequoia is finished in 2012, IBM believes the system will deliver 20 petaflops of performance using the company's next-generation BlueGene technology.

The only other petaflop supercomputer system listed on the Top 500 Supercomputer list is the Cray XT Jaguar supercomputer, which is located at the DOE's Oak Ridge National Laboratory in Tennessee.

However, before IBM builds the Sequoia supercomputer, the company's engineers will begin work on "Dawn," a high-performance computing (HPC) system, later this year. Dawn is based on IBM's BlueGene/P design and can deliver 500 teraflops, or 500 trillion calculations per second of performance.

Dawn will lay the ground work for Sequoia and IBM's new generation of BlueGene technology. IBM and DOE engineers are set to begin work on Sequoia in 2011 and finish the project a year later.

When both the Dawn and Sequoia systems are complete, the DOE is planning to use the two systems to connect the compute power of its laboratories at Los Alamos, Sandia and Livermore. This program will focus on various complex HPC simulations, including weapons research for the federal government. The system will also become part of the DOE's National Nuclear Security Administration Stockpile Stewardship program, which safeguards the nation's nuclear weapons.

Right now, IBM is not disclosing many details of its new BlueGene technology. However, the Sequoia supercomputer will be based on IBM's Power Architecture, which forms the basis on IBM's Power processors.

However, an IBM spokesman said that Sequoia will not be a hybrid system like Roadrunner. When IBM designed Roadrunner, engineers used a combination of IBM's own Cell processors as well as Advanced Micro Devices' Opteron chips to achieve the petaflop-level performance.

"The new technology will be based on future Blue Gene technology," Ron Favali, an IBM spokesman, wrote in an e-mail. "It is not a hybrid system like Roadrunner. The processor technology will be a future generation of IBM's Power processors."

In addition to processors based on IBM Power Architecture, the Sequoia supercomputer will use 1.6 petabytes of memory. The 96 refrigerated cabinets that make up Sequoia will hold 98,304 compute nodes and 1.6 million processing cores.

As of now, the fastest IBM BlueGene system is the BlueGene/L eServer Blue Gene Solution at the Lawrence Livermore labs. This supercomputer offers 478.2 teraflops of performance.