Method and data
One rented machine ran both codes on 2026-10-11. Everything the other pages quote or draw is in the files below.
One rented machine, both codes on it
| GPU (brae) | 1 × NVIDIA H100 PCIe 80 GB, plus 1 CPU core |
| CPU (OpenFOAM) | 26 cores of an Intel Xeon Platinum 8480+ (the instance's share; not a whole 56-core processor), 26 MPI ranks, scotch |
| Memory | 221 GiB |
| OpenFOAM | v2412, the packaged build; Open MPI 4.1.2 |
| brae | commit da7ef7e; its test suite 907 of 907 |
| Binary sha256 | 1d823cb400f6fdc957565d83b38e81557a86d6d345afe97dccddbd7ddf5d7ecb |
| OS | Ubuntu 22.04.5 |
| Measured | 2026-10-11; one OpenFOAM run and one brae run per setting |
interFoam shipped in release 0.5.0. The numbers on these pages were measured on commit da7ef7e, 27 commits ahead of release 0.5.0, and arrive with the next release.
Source of the experiment
K.M.T. Kleefsman, G. Fekken, A.E.P. Veldman, B. Iwanowski, B. Buchner, “A Volume-of-Fluid based simulation method for wave impact problems”, Journal of Computational Physics 206 (2005) 363–393. doi:10.1016/j.jcp.2004.12.007
Experiment carried out at MARIN; measurements from the SPHERIC benchmark archive, test 2 (test_case_2_exp_data.xls; test description by Issa and Violeau, release 1.1, 2006). spheric-sph.org/tests/test-02
The measured files
Nothing in them is estimated. A blank cell is a run that does not exist or a quantity a code does not print.
- kleefsman_speed.csvwhole-run wall clock, both codes
- vs_experiment_summary.csvevery gauge against the experiment
- vs_experiment_curves.csvevery gauge curve, 4 MB
- mesh_independence.csvevery quantity on every mesh
- mesh_gci.csvgrid convergence index
- conservation_vs_time.csvvolume, continuity, energy, momentum
- time_step_sensitivity.csvthree Courant limits
- interface_iou.csvoverlap of the water regions
- brae_small_co0.75.csvbrae, every time step, 0.4 M
- of26_small_co0.75.csvOpenFOAM, every time step, 0.4 M
- brae_vs_openfoam_42_tutorials_h100.csvthe 42 tutorials, 107 columns
- tutorial_solves.csvevery linear solve of the tutorials
- pinned_tolerance_gates.csvthe bound each tutorial is held to
- timing_repeatability.csvthe timing sweep run twice
- provenance.txthardware, versions, binary hash
- impact.csvP1's first impact in every run of the solver check, three meshes
- first_steps_4.89M.csvthe first eight time steps at 4.89 M, every run of the solver check