Back to Top
Skip to main content
NETL Logo
NETL and AI Partner Enhance Energy Modeling Techniques
NETL researcher Dirk Van Essendelft, Ph.D.

NETL researcher Dirk Van Essendelft, Ph.D.

NETL, in partnership with California-based Cerebras Systems Inc., is embracing new, efficient computer architecture that can accelerate research project simulations to make a clean energy economy a reality.

Cerebras is one of the world’s leading artificial intelligence (AI) hardware manufacturers. Working in conjunction with Cerebras, who designed the revolutionary wafer-scale engine (WSE) to tackle tough AI problems, NETL developed the Wafer-scale engine Field equation Application programming interface (WFA).

A programming interface allows different computer programs to communicate, and the WFA is enabling newer, more efficient means of generating simulation modeling data that will produce results faster while reducing the amount of energy consumed. The WFE is over 1,500-time more energy efficient per unit computing as compared to current state-of-the-art research computing resources.

NETL will be able to do complex simulations several hundred times faster so the Lab can design carbon capture equipment and novel energy generation and transfer equipment in less time to help accelerate the transition to a net zero economy.

NETL’s Dirk Van Essendelft, Ph.D., and his team have studied the WSE for several years, and have demonstrated that a WSE-enabled system running the WFA could help break down the memory and networking walls encountered during computational fluid dynamic (CFD) projects at the Lab.

CFD is the process of mathematically predicting fluid flow by solving the governing equations using computational power. CFD software and physical models of reacting multiphase flows whose purpose provide validated science-based modeling tools. In the context of energy, CFD is a valuable tool for analyzing fluid flow around turbines, cars, aircraft and other technologies. CFD has also become a useful tool in research data centers for analyzing thermal properties and modeling air flow in server rooms. CFD can help an administrator identify hot spots and learn where cold air is being wasted or air is mixing, thus establishing CFD as a fundamental discipline for advancing research on energy systems engineering.

To investigate the new WSE potential, Van Essendelft attended an AI science town hall in Late October 2019 and met Andy Hock, Cerebras vice president for product development at Cerebras. 

“He agreed that we should collaborate and put me in contact with Cerebras chief architect co-founder Michael James,” Van Essendelft said. “They agreed to put together a demonstration of the power of the WSE for high performance computing problems by writing the ‘Biconjugate Gradient Stabilized’ (BiCGSTAB) solver for their hardware and to benchmark it for us.” 

BiCGSTAB is the heart of Multiphase Flow with Interphase eXchanges (MFIX) where most of the time is consumed in computing CFD.

“The results were astoundingly fast, several hundred times faster,” Van Essendelft said. “Based on these results, NETL and Cerebras agreed to partner, and we began to make a general-purpose tensor algebra library that could support CFD on the WSE. This library eventually became known as the WFA. Applications for WFA include materials modeling, molecular dynamics and AI-accelerated scientific modeling, among others.”

NETL and Cerebras are now pursuing the first commercial application to run on the wafer scale engine through our interface. The impact will be transformative, improving modeling and simulation solution times by several hundred times over current methods.

“This means that work that currently takes a year to run using significant supercomputer resources can be done in a single day. If this work succeeds, it will be revolutionary,” he said. “We started development in spring of 2021 and did our first CFD simulation in December of 2022.  We now have a very capable library to solve a variety of scientific problems including materials problems and subsurface modeling in addition to CFD. Because of this success NETL and Cerebras are looking to develop an unstructured grid version of the WFA to support incredibly rapid modeling with complex geometries.”

NETL is a U.S. Department of Energy national laboratory that drives innovation and delivers technological solutions for an environmentally sustainable and prosperous energy future. By leveraging its world-class talent and research facilities, NETL is ensuring affordable, abundant and reliable energy that drives a robust economy and national security, while developing technologies to manage carbon across the full life cycle, enabling environmental sustainability for all Americans.