Cray building $600m exascale supercomputer to manage US nuclear stockpile

The exascale system will be based on Cray’s newly-announced Shasta architecture

The US Department of Energy (DOE) and National Nuclear Security Administration (NNSA) have announced they have signed a $600m deal with Cray to build the energy department’s first exascale supercomputer, dubbed “El Capitan”.

Exascale systems are capable of at least one exaFLOPs, or a billion billion calculations per second. El Capitan, which will have a peak performance of more than 1.5 exaflops when delivered in 2022, will be used to run applications that safeguard the safety, security and effectiveness of the US’s nuclear stockpile, and to evaluate evolving threats to the country’s national security.

The supercomputer is expected to run nuclear applications at 50 times the speed of the computer currently tasked with the job, the Lawrence Livermore National Laboratory (LLNL) Sierra system, currently the world’s second most powerful computer with 125 petaflops of peak performance.

In addition to sizable performance gains, El Capitan is projected to be at least four times more energy-efficient than its predecessor. The system will be the DOE’s third exascale-class supercomputer, alongside Argonne National Laboratory’s “Aurora” and Oak Ridge National Laboratory’s “Frontier” system – both also built by Cray.

“NNSA is modernizing the Nuclear Security Enterprise to face 21st- century threats,” said Lisa E. Gordon-Hagerty, DOE under secretary for nuclear security and NNSA administrator. “El Capitan will allow us to be more responsive, innovative and forward-thinking when it comes to maintaining a nuclear deterrent that is second to none in a rapidly-evolving threat environment.”

The system will utilise Cray’s newly-announced Shasta architecture, an open software platform built for data-intensive workloads, such as simulation or those needed to train large machine learning models.

Shasta leverages Kubernetes container orchestration to enable converged high-performance computing (HPC) and AI workflows and includes standardised and supported APIs for integration, data access and software ecosystem extensibility and interoperability. It also comes with an analytics software stack purpose-built for exascale computing.

Cray and LLNL will work together in the coming months to decide which processor and GPU components the system will use.

“We are honoured to be a part of this historic moment to deliver the next U.S. exascale supercomputing system to the DOE, NNSA and LLNL in support of their incredibly important mission,” said Pete Ungaro, president and CEO of Cray.

“We couldn’t be more excited that Cray’s Shasta systems, software and Slingshot interconnect will be the foundation for the first three U.S. exascale systems. El Capitan will incorporate foundational new software technologies from Cray that are critical for the exascale era where digital transformation and the convergence of modelling simulation, analytics and AI are driving new, data-intensive workloads at extreme scale.”