New Sandia supercomputer aims to sharpen simulations safeguarding nuclear arsenal

Sandia National Laboratories welcomed a new supercomputer this month, one that could reshape simulations and ensure the reliability and safety of the U.S. nuclear stockpile.

The prototype system, named Spectra, was built using specialized chips that can analyze code to prioritize tasks in real-time.

“It’s a completely new architecture; it’s a completely new way of doing things,” said James Laros, Sandia senior scientist and project lead. “That’s what the real challenge is — to take what’s already complicated and try to run it on a new architecture that’s never been used before, but has the potential to be superior in some dimension.”

The news comes on the heels of Los Alamos National Laboratory announcing the selection of HPE and Nvidia Corp. to build two new supercomputers, named Mission and Vision, expected to be operational by 2027.

Spectra is the first supercomputer to use the new chip architecture, designed by Israel-based NextSilicon. Researchers will test how the system handles national security-related tasks like advanced fluid dynamics simulations, which assess the safety and reliability of nuclear deterrents — think weapons or communications — without underground testing.

Configurations using NextSilicon’s Maverick-2 dual-die accelerators have the potential to have higher performance compared to other technologies. It may also complete tasks using much less power than usual, something Laros said could be a “big win” as access to energy becomes a growing problem.

Laros said Spectra was installed and came online Nov. 1, joining about a dozen other Sandia supercomputers. The system is the second platform involved in the lab’s Vanguard program, an initiative exploring emerging technologies used in advanced simulation and computing applications.

The Vanguard program, overseen by the National Nuclear Security Administration, includes work from the nation’s three nuclear weapons labs: Sandia, LANL and Lawrence Livermore National Laboratory in California.

“There’s all kinds of things that happen in that space. Everything from the actual nuclear weapon to the vehicles, each laboratory has a piece of that pie,” Laros said. “We don’t do this just because it’s good for Sandia — we do it because it’s good for the complex.”

The program’s first supercomputer, named Astra, was the world’s fastest ARM-based system in 2018. Laros said ARM-based chips have been “ubiquitous in the world” for a long time, used in everyday items like cell phones and automobiles, but were never used for high-performance computing.

Astra’s success would pave the way for LANL to install Venado, a partially ARM-based supercomputer built to prioritize performance and workflow efficiency, in 2024.

Currently, Spectra is in its prototyping phase before it can become a “formal system,” Laros said, which can be considered a risky endeavor. Project timelines for advanced architectures like this can span anywhere from six months to a couple of years, he said.

But continuing to build and take a chance on new advanced systems is imperative to Sandia’s mission to protect the U.S. nuclear stockpile, Laros said.

“Let’s say you only had one technology. It could be the greatest technology in the world, and it could do our mission perfectly — but what if it just went away? Then what do we do?” Laros questioned. “We can’t just have one.”