Imagine storing the equivalent of 500,000 HD movies worth of data—all dedicated to simulating how the universe evolved from the Big Bang to today. That's exactly what an international team of researchers has just released through the FLAMINGO project, and it represents one of the most ambitious computational cosmology efforts in history.
The simulations consumed more than 50 million CPU hours on the COSMA 8 supercomputer at Durham University. To put that in perspective, one of the flagship runs took 42 days running on roughly 30,000 CPUs simultaneously. If you tried that on a single high-end processor, you'd be waiting for centuries.
FLAMINGO—which stands for "Full-hydro Large-scale structure simulations with All-sky Mapping for the Interpretation of Next Generation Observations"—is led by researchers from Leiden University and the Virgo Consortium. What makes it special isn't just scale, though that's impressive enough. The largest simulation box spans 2.8 gigaparsecs, about 9 billion light-years on a side, tracking up to 300 billion particles.
The real breakthrough is what those particles represent. Traditional cosmological simulations often track only dark matter, treating the ordinary matter we're made of as an afterthought. FLAMINGO includes full hydrodynamics: stars, galaxies, gas, cooling, star formation, supernovae, and active galactic nucleus feedback. It even models massive neutrinos as explicit particles using what's called the "δf method" to minimize computational noise, alongside dark matter and dark energy.
This matters because baryonic effects—the behavior of ordinary matter—can suppress the matter power spectrum by up to 20%. Ignore that, and your predictions about cosmic structure formation drift away from reality.
The team calibrated their models using machine learning techniques, specifically Gaussian process emulation, matching simulations to observed galaxy stellar mass functions and cluster gas fractions. This is where computational cosmology meets modern AI—not to hallucinate results, but to efficiently explore parameter space and find physical models that match what telescopes actually see.
Why does this matter beyond bragging rights about data volume? Because we're living through a golden age of observational cosmology. The is peering back to the earliest galaxies. Euclid is mapping billions of galaxies across cosmic time. DESI is measuring the expansion history with unprecedented precision. LSST will come online soon, detecting billions more objects.
