Lawrence Livermore National Laboratory has broken ground on its Exascale Computing Facility modernization project to provide advanced supercomputer services through expansion on the campus.
The project aims to significantly upgrade the mechanical and electrical capabilities of the lab's Livermore Computing Center.
Since 2004, LLNL has had some of the largest, fastest and most advanced systems, and to further this venture, National Nuclear Security Administration (NNSA) facilities require high performance machines to exceed current capacities, according to lab officials.
As a result, construction to expand system capacity was advanced with new computer designs, increased requirements, number of processors per system, and the density of processors per rack.
The project, which did not have a formal groundbreaking ceremony because of COVID-19, involves expanding the current cooling system from 10,000 tons to 28,000 tons by adding piping and pumps that will add cooling tower capacity. Currently, the electrical system will also be upgraded from 45 megawatts to 85 megawatts to ensure high-performance computing.
"Since then we've been housing high performance computing machines in that facility and that particular building has ample square footage," said Anna Marie Bailey, the program facility manager. "We have nearly 50,000 square feet of data center space and we're only adding the power to meet the needs of exascale computing in the future."
LLNL officials estimate that the modernization project will reach completion in July 2022, but it will be substantially completed by November 2022 so crews can begin preparation for a separate infrastructure package. One of the systems being used to showcase facility's new capabilities is known as El Capitan. It is projected that this supercomputer will be the world's fastest supercomputer by 2023.
According to LLNL, El Capitan will run roughly 10 times faster on average compared to the facility's current Sierra system. Bailey stated that each machine lasts on the floor for five to seven years, and are eventually retired for the next effective supercomputer.
"It's not a simple thing. A lot of this equipment can take up to a year to 18 months to manufacture," she said. "The laboratory is dedicated to that strategic planning and so in my role as the HPC engineer, I have to look at that whole gamma of what's coming down the line, what type of systems do we forecast ... and try to forecast what types of technologies will provide over several generations."