Supercomputer Taps 15,000 GPUs To Rebuild The Early Universe
When trying to model the birth of the universe, computational cosmologists are usually forced to choose between simulating vast volumes of space at a coarse resolution or zooming into tiny patches with intricate detail. Lumina changes all that: spanning a cubical region measuring roughly 1.6 billion light-years across, the simulation tracks ordinary matter, dark matter, and radiation across all that space while retaining enough spatial resolution to model individual galaxies, stellar nurseries, and their surrounding gas flows.
Bringing this map life was an international research team that utilized the Frontier supercomputer located at Oak Ridge National Laboratory in Tennessee. Harnessing more than 15,000 GPUs and 100,000 CPU cores, the simulation begins roughly 45 million years after the Big Bang, in the dark ages. Here, gravity slowly pulls tiny fluctuations in matter together, forming thick over-densities for the first generation of galaxies and stellar clusters.
Lumina is currently open to the global scientific community, and could prove an indispensable sandbox for upcoming research whereby users can now pitch the model's predictions against incoming telescope data, for example. Instruments like the JWST and Nancy Grace Roman Telescope routinely peer deep into the past, returning images of ancient galaxies and quasars that frequently challenge our theoretical limits. However, telescopes only capture scattered, frozen snapshots of specific moments. Lumina stitches those disparate observations into a coherent, physics-driven and visual narrative.
