Lawrence Livermore National Laboratory, which has pioneered the use of high performance computing in science since the 1950s, dedicated its most powerful computer last week.
Sierra, as the computer is named, is said to be capable of running at a speed of 125 million billion operations per second, or 125 petaflops. Sierra’s first components were delivered four years ago, but the completed system is only now approaching full capacity.
Early next year, Sierra will be switched to classified operations to model nuclear weapons performance in such detail that the programs could take hundreds of hours to run even on the world’s fastest computers.
These so-called multi-physics codes integrate efforts by teams from around the laboratory, requiring “from tens of processors to tens of thousands of processors to hundreds of thousands of processors,” according to Chris Clouse, associate program director of computational physics at LLNL.
Although it is located in Livermore, Sierra is a facility of the National Nuclear Security Administration, or NNSA.
As such, weapons physicists from all three weapons laboratories -- Los Alamos, Sandia and Lawrence Livermore National Laboratories -- “have equal rights” to its use, according to Mike McCoy, LLNL’s program director for advanced simulation and computing.
Sharing the use of large, expensive facilities is common practice throughout the nuclear weapons complex, both for cost and security reasons. Researchers doing classified experiments and calculations must protect their data and typically cannot use academic or commercial resources.
Sierra is some 6 times more powerful than LLNL’s next fastest machine, Sequoia. It is the product of several major players in the computer world including IBM, NVIDIA and Melanox.
Reflecting the team approach to developing, installing and bringing the computer up to full capacity, LLNL director Bill Goldstein issued a statement thanking “everyone involved in getting us to this point: our sponsors at NNSA, our industry and national lab partners and our own dedicated staff.”
Role of Computers
LLNL pioneered the use of computers in nuclear weapon design and in other fields of science beginning in the 1950s. Much of the development of early high-powered computer systems was carried out in response to Livermore’s needs, under the support of the old Atomic Energy Commission.
That is no longer true, however. Today, data science, cloud computing and entertainment graphics are the lifeblood of the industry, and computer companies are intensely competitive in pursuing innovation.
Government supported labs like LLNL draw on those innovations, working in close partnership with the companies to “tailor what they are building for high performance scientific computing,” according to LLNL’s Chris Clouse.
He sees NNSA as the only agency seriously promoting high performance computing today.
The arrangement of Sierra processing units is a case in point. Computer industry publications are calling Sierra the first major NNSA computer to use what is known as heterogeneous architecture, referring to the coordinated use of CPUs -- the traditional central processing units of everyday computers – with GPUs, graphical processing units which can greatly speed certain repetitive computations.
Both Clouse and McCoy see heterogeneous architecture as helping to define the path toward still more powerful computers in the future.
Sierra is the second in a four-computer NNSA sequence intended to lead the U.S. into the next faster realm of supercomputing. This is commonly called “exascale” computing, signifying calculations at a rate of at least a billion billion operations per second, a thousand times faster than petascale.
The first NNSA computer to reach that goal may be a follow-on machine at LLNL called El Capitan, now scheduled for operation in 2023.
El Capitan may not be first, however.
Two facilities of a different branch of the U.S. Department of Energy, the Office of Science, plan to demonstrate exascale computers earlier. These are Argonne National Laboratory in 2021 and Oak Ridge National Laboratory in 2022.
Nor will the first exascale computer necessarily be American. The Japanese and Chinese have extremely capable computer programs, and both have declared their intention to dominate the field.