Cerebras Systems, known for its pioneering work in wafer-scale processing, has partnered with researchers from Sandia, Lawrence Livermore, and Los Alamos National Laboratories to achieve a major milestone in molecular dynamics (MD) simulations. For the first time in the field’s history, the team reached over 1 million simulation steps per second. A single Cerebras Wafer Scale Engine accomplished more than 1.1 million steps per second, a staggering 748 times faster than the world’s top supercomputer, ‘Frontier.’

Additionally, Cerebras surpassed Anton 3, the leading purpose-built supercomputer for molecular dynamics. While Anton 3 relies on 512 specialized processors and 400 kilowatts of power, the Cerebras system uses only a single accelerator, consumes just 7% of the power, and outperforms Anton 3 by 20%.

This breakthrough allows scientists to complete the equivalent of two years’ worth of GPU-based simulation work in just one day, significantly accelerating innovation in molecular simulations. It also offers new insights into material structure and function. As the technology extends to biomolecules, it could unlock advancements in protein folding, medicine, and drug development.

Supercomputing performance can be achieved through weak or strong scaling. Weak scaling involves increasing the size of a simulation, while strong scaling focuses on speeding it up. While many supercomputers use multiple GPUs to scale simulations, Cerebras achieves strong scaling, drastically accelerating simulation speed. This makes it possible for scientists to simulate materials over much longer periods, essentially allowing them to peer into the future of these materials.

Traditional supercomputers are limited to simulating materials at the atomic level for just a few microseconds. However, Cerebras’ breakthrough enables simulations at the millisecond timescale, offering a view that is 1,000 times more advanced than what is currently possible.

The collaboration between the NNSA laboratories and Cerebras is part of the Advanced Memory Technology (AMT) program, which aims to enhance exascale supercomputers by 40 times by 2025. Thanks to Cerebras’ deployed wafer-scale systems, the teams achieved a materials science breakthrough that exceeded the AMT program’s speedup goal by over four times.

With long timescale simulations now possible, scientists can explore a variety of previously unreachable phenomena:

• Materials scientists can now examine the long-term behavior of complex materials, such as how grain boundaries evolve in metals, leading to stronger, more resilient materials.
• Pharmaceutical researchers can simulate protein folding and drug-target interactions over realistic timescales, accelerating the development of life-saving therapies.
• Renewable energy experts can optimize catalytic reactions and design more efficient energy storage systems by simulating atomic-scale processes over extended periods.

In addition, Sandia recently deployed a cutting-edge Cerebras CS-3 testbed for future AI workloads, further solidifying the company’s role in advancing the field.