High-performance artificial intelligence computing startup Cerebras Systems Inc. announced a key milestone today, saying it has carried out the first-ever simulation of a high-resolution convection workload at near real-time rates.
The simulation was performed using the WSE Field-equation application programming interface, or WFA, that was developed by the U.S. Department of Energy’s National Energy Technology Laboratory. It could run several hundred times faster than what is possible using traditional distributed computers.
The demonstration was carried out by the Neocortex AI supercomputer at the Pittsburgh Supercomputing Center. The Neocortex AI Supercomputer is powered by multiple Cerebras CS-2 appliances. Those are extremely powerful AI compute platforms incorporating the company’s WSE-2 AI chip. They’re based on a so-called wafer-scale architecture that’s optimized for AI workloads.
A single WSE-2 AI chip is said to contain 2.6 trillion transistors, or 2.55 trillion more than the most advanced graphics processing unit on the market. The CS-2 system combines a single WSE-2 chip with cooling equipment, networking hardware and other supporting components, making it easier to deploy by removing the need for customers to install the necessary supporting components on their own.
Cerebras said it partnered with NETL and PSC to simulate “thermally driven fluid flows,” also known as natural convection, an application of computational fluid dynamics. These fluid flows are natural occurrences that cause high winds, lake effect snowstorms and tectonic plate motion.
The simulation was made up of about 200 million cells and was focused on a phenomenon called “Rayleigh-Bénard” convection, which occurs when a fluid is heated from below and cooled from above. In nature, this phenomenon can result in severe weather such as downbursts, microbursts and derechos, which can cause hurricane and tornado-force winds, heavy rains and flash floods. Rayleigh-Bénard is also believed to be the driving force behind magma movement in the Earth’s core, as well as plasma movement within the sun.
Cerebras said the WFA API powered by its CS-2 system simulated Rayleigh-Bénard convection about 470 times faster than what was possible on NETL’s existing Joule Supercomputer, which is powered by traditional GPUs.
NETL Lab Director Brian J. Anderson said simulating natural convection in real time will enable researchers to accelerate and improve the design process of projects vital to mitigating climate change and securing clean energy, such as carbon sequestration and blue hydrogen production.
“This simulation wouldn’t have been possible without Cerebras’ industry-leading CS-2 system, which is the main compute element of PSC’s Neocortex supercomputer,” he said. “Running on a conventional supercomputer, this workload is several hundred times slower, which eliminates the possibility of real-time rates or extremely high-resolution flows.”
Paola Buitrago, PSC’s principal investigator and director of AI and big data, said the Neocortex AI supercomputer was introduced in June 2020 to advance interactive AI for the research community. “This recent achievement, in partnership with NETL and Cerebras Systems, is one of many great discoveries we have been aiming to achieve to advance fundamental research across scientific workloads,” she added.