World’s fastest supercomputer runs record-breaking fluid simulation for rocket testing

Five hundred quadrillion degrees of freedom. That's the staggering computational scale researchers achieved using Lawrence Livermore's El Capitan supercomputer to simulate rocket exhaust plumes in unprecedented detail. The team tackled one of aerospace engineering's thorniest problems: what happens when dozens of rocket engines fire simultaneously, creating turbulent interactions that can send searing gases back toward the vehicle.

Their simulation modeled the complex exhaust dynamics of a configuration inspired by SpaceX's Super Heavy booster, using all 11,136 nodes and more than 44,500 processing units. The breakthrough came through a new technique called Information Geometric Regularization, delivering an 80-fold speedup over previous methods while slashing memory use and energy consumption.

"The method is faster and simpler, uses less energy on El Capitan, and can simulate much larger problems than prior state-of-the-art — orders of magnitude larger," said Georgia Tech's Spencer Bryngelson, the project's lead.

As private spaceflight increasingly relies on arrays of compact engines rather than massive single boosters, these simulations could replace costly physical experiments with predictive modeling, potentially revolutionizing rocket design.