An exascale many-physics flow solver for compressible multi-phase simulations, scaling to 200 trillion grid points on 43K+ GPUs.
MFC is an exascale many-physics flow solver for simulating compressible multi-phase flows at massive scales. It solves complex fluid dynamics problems involving shocks, interfaces, phase change, and bubble dynamics, scaling efficiently to hundreds of thousands of GPUs on the world's largest supercomputers. The project addresses the need for open, high-performance CFD tools that can leverage modern GPU architectures for groundbreaking scientific simulations.
Computational physicists, fluid dynamics researchers, and engineers working on large-scale simulations of multi-phase flows, shock physics, or bubble dynamics. It is also suited for HPC developers and institutions needing portable, GPU-accelerated CFD solvers for exascale systems.
Developers choose MFC for its proven exascale scalability, compact and maintainable codebase, and comprehensive built-in multi-phase physics. Its MIT license, active community, and support for both NVIDIA and AMD GPUs offer a unique open-source alternative to proprietary high-performance CFD software.
Exascale multiphase flow solver — 2025 Gordon Bell Prize Finalist | 200T grid points on 43K+ GPUs
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Achieves ideal weak scaling to over 43,000 GPUs on supercomputers like El Capitan and Frontier, with near compute-roofline performance, as validated by its Gordon Bell Prize finalist status.
At ~40,000 lines of Fortran enhanced with Fypp metaprogramming, it balances readability with high-performance, making it easier to audit and modify than million-line codebases.
Natively supports 4, 5, and 6-equation models, phase change, surface tension, and bubble dynamics, eliminating the need for external libraries for core advanced features.
Runs on NVIDIA and AMD GPUs, CPUs, and offers Docker, Codespaces, and Homebrew installs, enabling deployment from laptops to exascale clusters with minimal friction.
Development requires expertise in Fortran and Fypp, which can be a steep learning curve compared to solvers using more modern languages like C++ or Python, limiting contributor accessibility.
Setting up MFC on custom HPC systems involves navigating 16+ system templates and multiple compilers, which the README admits can be challenging despite portability claims.
Relies entirely on command-line tools and external software like ParaView for visualization, lacking built-in graphical interfaces that some commercial CFD suites offer for easier workflow integration.