A synthetic observation of the simulated galaxy in blue stellar light and warm dust

PHEASANT Collaboration

Building a Milky Way,Star by Star.

AI-accelerated exascale simulations of galaxy formation and evolution while resolving individual stars.

Performance at a glance

Exascale Simulations
of Our Galaxy.

PHEASANT brings the lives of individual stars into a galaxy-scale experiment, running at the scale of exascale systems.

5.4Trillion

Particles

976PFLOP/s

Perfomance of the Gravity Kernel

144Billion

Particles Processed second

0.22Solar Mass

Mass Resolution

One simulation, many scales

From spiral arms
to single stars.

Move through a continuous numerical universe—from a hundred thousand parsecs (~a couple of hundred thousand light-years) down to stellar-scale events that reshape stellar nurseries.

A simulated barred spiral galaxy seen from above10 kpc x 10 kpc

The whole system

Galaxy

Follow spiral structure, stellar migration, and the flow of matter across an entire Milky Way-mass galaxy.

Simulated star-forming regions along the galactic bar500 pc x 500 pc

Where stars are born

Star-forming regions on a bar

Resolve turbulent gas clouds, superbubbles, and the clustered environments that turn cold gas into stars.

A young stellar association in the simulated galaxy100 pc x 100 pc

Every source matters

Young stellar associations

Model individual stellar evolution and the supernovae that return energy and newly forged chemical elements to the galaxy.

Technologies behind PHEASANT

Three breakthroughs.
One Milky Way.

01Milky Way-mass simulation resolved into individual stellar populations

A Milky Way, star by star

We achieved subsolar-mass resolution in a Milky Way–mass simulation, resolving individual stars and following their evolution within one continuous model of the galaxy.

02Detailed multiphase structures computed by the exascale simulation

Portable Performance at Exascale

Our CUDA and SYCL GPU implementations, together with optimized x86 and Arm CPU implementations, sustain high performance across NVIDIA, AMD, and Intel GPUs as well as x86 and Arm CPUs, enabling highly scalable multi-scale physical simulations on diverse exascale systems.

03Individual stars and supernova feedback in a forty-parsec region

AI Surrogate Modeling for Supernova Feedback

An AI surrogate model replaces computationally expensive and short-timescale simulations of supernova feedback, removing a major bottleneck in multi-scale galaxy simulations while preserving its impact on the matter cycle and galactic evolution.

Featured movie

Watch a galaxy
come alive.

View all visualizations
A face-on view of the PHEASANT simulated galaxyJourney Through a Star-by-Star GalaxyFeatured film · 00:24Watch the movie

Research & collaboration

Built to make the
invisible visible.

Publications

Methods, science results, and release notes will be collected here as the project develops.

Coming soon

Team

PHEASANT is built by a collaboration spanning astrophysics, deep learning, and high-performance computing.

Meet the collaboration