NETL and Cerebras Look to Wafer Scale Engine to Increase Onsite Energy Efficiency

NETL and project partner Cerebras are advancing high-performance computing (HPC) and artificial intelligence (AI)-physics model coupling that can greatly reduce energy consumption and greenhouse gas emissions associated with energy research projects by as much as 30%.

"Science and engineering applications of national importance will require increasingly capable advanced computing to model complex phenomena and manage extensive amounts of data," said Tammie Borders, the associate director for computational science and engineering at NETL. "While computing capabilities have continued to double every two to two and a half years, power efficiency, on the other hand, has not kept pace. This means disproportionately increasing costs, power consumption and carbon footprint."

A potential solution to this power efficiency problem lies in the Wafer-Scale Engine (WSE) developed by Cerebras Systems, which has essentially taken the power of a supercomputer and reduced it down to a single chip. The equivalent of each computational node has been shrunk down to just a couple hundred microns in size.

"Research computing represents about half of NETL's total energy consumption and CO₂ emissions, which is representative of data center and computing growth worldwide," Borders said.

"Moving half of the Lab's projected research computing to the WSE could lead to a reduction of approximately 30% of on-site emissions. This move would be the largest and least expensive single-source emission reduction NETL could make to reduce its onsite carbon footprint. As a leading developer of efficient energy technology, we can make the research and development of said technology more and efficient too. Think of it as practicing what we preach by embracing the newest tools at our disposal."

NETL and Cerebras have been developing HPC and HPC/AI applications for the WSE since 2019 and have demonstrated an impressive set of benchmark records. The first collaborative work was to evaluate the biconjugate gradient stabilized performance on a structured grid, as it is the heart of NETL's Multiphase Flow with Interphase Exchange (MFiX) CFD software.

Cerebras did the WSE programming and NETL did competitive benchmarks with the solvers in MFiX on Joule 2.0. After seeing such impressive speed gains, NETL made the decision to start developing an Application Programming Interface (API) called the WSE Field-equation API to enable researchers to easily program the WSE.

Cerebras is a private, U.S.-owned computing chip company with a global footprint. It was founded in 2016 and to date has raised over $700 million and was valued at $4 billion in 2021. As a team of pioneering computer architects, system, software and machine learning (ML) engineers, Cerebras is pioneering a new class of computer system designed from first principles to meet today's ML and scientific computing challenges.

NETL is a U.S. Department of Energy national laboratory that drives innovation and delivers 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.

Public link

Please use the above public link if you want to share this noodl on another website.