Sandia National Laboratories, which runs simulations of nuclear warhead behavior and hypersonic weapons trajectories, is reassessing its chip suppliers as Nvidia and Advanced Micro Devices focus more heavily on Artificial Intelligence. For more than a decade, that work relied on chips from Nvidia ($NVDA) and Advanced Micro Devices ($AMD). Now Sandia is evaluating chips from an Israeli startup for that mission-critical work instead.
The main concern is double-precision floating point computation, a specification needed to process very large and very small numbers in the same calculation without losing accuracy to rounding errors. Nvidia and AMD previously competed aggressively in this area to win supercomputing contracts from universities and government labs, but Artificial Intelligence workloads do not require the same level of double-precision performance. Per Reuters, the double-precision performance of Nvidia’s forthcoming Rubin chips has declined by some measures, raising concern across the high-performance computing industry. AMD is releasing a version of its chips aimed at scientific computing, but Sandia also faces supply chain pressure as it weighs its options.
NextSilicon is offering a different approach. Its chips use a data flow architecture rather than the graphics processing units and central processing units that dominate the market, and the design is intended to reprogram itself in real time to run more efficiently on each task. According to Reuters, this architecture reduces the time and energy spent moving data between the processor and system memory. The chips also support double-precision computation, combining energy efficiency with the scientific precision that mainstream Artificial Intelligence-focused chips have begun to deprioritize.
This week, Sandia, NextSilicon, and Penguin Solutions announced that the systems passed a key technical milestone after clearing a battery of general supercomputing benchmarks. That result puts the chips in contention for government use and sets up a decision this fall on whether to move to more demanding tests that more closely resemble the nuclear security calculations the systems would eventually handle. Sandia says it must keep alternative options available because its mission cannot be deferred.
Sandia’s evaluations often influence the broader chip industry. The lab began pushing Intel, AMD, and Nvidia to develop liquid cooling for chips more than a decade ago, when the technology was considered experimental. Liquid cooling is now standard across the industry. If NextSilicon advances through Sandia’s process, the endorsement could help accelerate adoption beyond government labs, even as Nvidia says it is still working toward chips that can handle both Artificial Intelligence and scientific workloads.
