Revolutionary Rocket Simulation Achieves Exascale Performance on El Capitan

Researchers have accomplished a groundbreaking feat by using Lawrence Livermore National Laboratory’s (LLNL) exascale supercomputer, El Capitan, to run the largest fluid dynamics simulation to date. This simulation exceeded one quadrillion degrees of freedom within a single computational fluid dynamics (CFD) problem, focusing on the interactions between multiple rocket plumes.

El Capitan, supported by the U.S. National Nuclear Security Administration’s (NNSA) Advanced Simulation and Computing (ASC) program, facilitated this achievement before shifting to classified operations earlier this year. The project was led by Georgia Tech scientists, with collaboration from AMD, NVIDIA, HPE, Oak Ridge National Laboratory, and New York University.

This work is a finalist for the 2025 ACM Gordon Bell Prize, the top honor in high-performance computing. The winner will be announced at the SC25 conference in St. Louis on November 20. To simulate the turbulent exhaust flow from multiple rocket engines firing simultaneously, the team employed a novel shock-regularization technique called Information Geometric Regularization (IGR), developed by Bryngelson, NYU’s Schäfer, and Cao.

Using over 11,000 nodes and more than 44,500 AMD Instinct MI300A APUs on El Capitan, they achieved a simulation with over 500 trillion grid points. This was extended to the Frontier supercomputer at Oak Ridge National Laboratory, surpassing one quadrillion degrees of freedom. The simulations, run with the open-source MFC code, modeled the full exhaust dynamics of a scenario inspired by SpaceX’s Super Heavy booster.

This simulation sets a new benchmark in exascale CFD performance, enhancing the ability to predict rocket behaviors and design more efficiently. It moves the industry toward replacing costly physical tests with highly detailed computational modeling.

Georgia Tech’s Bryngelson highlighted the significance of the achievement. “This marks a major step forward in fluid dynamics,” he said. “The new method is faster, simpler, and consumes less energy, enabling us to simulate problems far larger than before.”

The project also served as a system stress test for El Capitan, helping LLNL evaluate its scalability and readiness before the machine’s official classified deployment. Livermore’s team reviewed the system’s full capabilities, demonstrating its potential for supporting critical national security missions.

FAQs

Q: What is the significance of this simulation?
A: It demonstrates the ability to perform extremely detailed CFD at exascale, advancing rocket design and reducing reliance on physical testing.

Q: What techniques were used to achieve this simulation?
A: The team employed a novel shock-regularization method called Information Geometric Regularization and utilized massive parallel computing resources.

Q: Why is this important for aerospace engineering?
A: It allows for accurate prediction of complex exhaust interactions, leading to safer, more efficient rocket designs and breakthroughs in propulsion technology.

Q: What is the Gordon Bell Prize?
A: It is a prestigious award recognizing outstanding achievements in high-performance computing. This project was a finalist for the 2025 award.