An Israeli startup called NextSilicon built a supercomputer, Spectra, using its novel “Maverick-2” chips.

This system just passed real-world testing by the US-based Sandia National Laboratories, proving it can handle highly complex national security and scientific workloads.

The achievement is the second ever for the US government’s elite Vanguard program. Vanguard exists to hunt down wild, unproven chip architectures and test them against real-world national security workloads. 

It does not hand out participation trophies. To pass, Spectra had to run highly complex nuclear stockpile and molecular physics simulations without crashing, dragging, or fracturing code.

It passed with flying colors. This successful milestone establishes the new chip technology as a strong candidate for future, large-scale government supercomputer deployments.

“System acceptance at Sandia is not a checkbox. It means Maverick-2 ran mission-relevant workloads, demonstrated system stability, and showed the computational scientists at Sandia what this architecture can deliver,” said Elad Raz, Founder and CEO of NextSilicon. 

“This is a significant step toward what we have been building: an accelerator that delivers performance while reducing power consumption. For HPC organizations evaluating next-generation infrastructure, Spectra begins to show what Maverick-2 can do when put to the test,” Raz added.

High-performance computing

Despite dominating AI and supercomputing data centers with their unmatched raw power, Nvidia‘s massive GPUs have drawbacks. These are expensive architectures that require substantial electricity, ultimately placing a heavy strain on local power grids.

Built through a partnership between Sandia National Laboratories, NextSilicon, and Penguin Solutions, the Spectra supercomputer features 64 compute nodes and 128 Maverick-2 dual-die accelerators. 

The Maverick-2 sets a new efficiency standard by delivering over 4x the performance-per-watt of standard GPUs and 20x that of high-end CPUs, while cutting total ownership costs in half. 

Hence, it provides sustainable, high-performance computing that helps data centers and researchers meet environmental goals without sacrificing speed.

To pass the rigorous Vanguard evaluation program, Spectra successfully ran demanding simulations like HPCG, LAMMPS, and SPARTA.

Compared to standard GPUs, NextSilicon’s Maverick-2 chip adapts its hardware resources to each application at runtime, demonstrating the performance and stability required to qualify for the elite program.

Vanguard’s first system, Astra, proved in 2018 that low-power Arm processors could scale up to run supercomputers. Today, Arm architecture is a pillar of modern data centers. Spectra aims to repeat that history.

“The Vanguard program exists to put new architectures through rigorous evaluation against workloads that are directly relevant to our mission,” said James H. Laros III, Senior Scientist and Vanguard Program Lead at Sandia National Laboratories. 

“Our partnership succeeded in executing all benchmarks and applications specified, meeting the acceptance criteria we defined for this phase in the program. This outcome is exactly what our process is designed to test, and it gives us a solid basis for continued evaluation of this technology,” he added. 

Future deployment

Already deployed at dozens of customer sites worldwide, Maverick-2 has expanded its footprint into one of the US government’s most heavily scrutinized computing programs following its success at Sandia. 

“For commercial HPC buyers assessing alternatives to GPU-based infrastructure, Spectra offers a reference point grounded in rigorous evaluation against mission-critical workloads in an operational national security environment,” the company stated. 

The development proves that tomorrow’s computing breakthroughs might not come from building bigger, hotter chips.