ALBUQUERQUE, New Mexico - In a nondescript building on Kirtland Air Force Base, liquid-cooled supercomputers hum through complex math problems for the U.S. government, simulating hypersonic nuclear weapons.
For over a decade, chips from Nvidia or AMD handled this demanding work. But as these firms focus on AI, supply shortages worry Sandia National Laboratories.
Steve Monk, Sandia’s high-performance computing manager, said the pressure from the computing front and supply chain is stressful.
This opens the market to newcomers like NextSilicon, an Israeli startup whose chips use a data flow architecture for efficiency. Sandia is testing them, with a decision expected this fall.
Double-precision computing, crucial for scientific work, is declining in Nvidia's new Rubin chips, pushing labs to explore alternatives. Sandia aims to ensure chip availability regardless of market shifts.
