Mallard
High-order unstructured finite-volume CFD solver for compressible Euler, Navier-Stokes and reacting flows. It runs on CPUs and on NVIDIA and AMD GPUs via Kokkos.
Release timeline (newest first, grouped by year and version)
2026 10 entries
0.x
- Mallard 0.10.0
v0.10.0releaseDevice-side step scheduling and CUDA graphs for every kind of run, including reacting flow and large-eddy simulation. Each run writes a self-contained report.html, and new post, info, diff and pack commands plus a Python package work on saved output.
source: github-release - Mallard 0.9.1
v0.9.1releaseMallard 0.9.1 fixes two problems in 0.9.0. Viscous [[forces]] were slightly wrong on several ranks, and the release published no container images. Inputs and results are otherwise unchanged from 0.9.0. Fixed - [[forces]] on several ranks exchanges the halo before computing the wa
source: github-release - Mallard 0.9.0
v0.9.0releaseMallard 0.9.0 makes regions a single concept used everywhere. Points, curves, surfaces and volumes, defined once and named, drive sources, sponges, initial states, probes, snapshots and curved boundaries. It adds probes of every dimension with exact fluxes, explicit boundary-cond
source: github-release - Mallard 0.8.0
v0.8.0releaseMallard 0.8.0 adds curved high-order boundaries, which bring TENO-E orders 3–6 to their design accuracy at curved walls, where straight-sided faces did not converge. It sets up 3D TENO 6–9.5× faster on the GPU and fits about 15% more cells on one (twice as many with opt-in single
source: github-release - Mallard 0.7.0
v0.7.0releaseMallard 0.7.0 adds large-eddy simulation with explicit subgrid-scale and combustion models and a kinetic-energy-preserving hybrid flux, which also brings the Re_τ = 180 channel DNS within 2% of Moser, Kim & Mansour. It adds synthetic turbulent inflow, makes 3D TENO 2.2× faster an
source: github-release - Mallard 0.6.0
v0.6.0releaseMallard 0.6.0 adds axisymmetric flows, non-reflecting boundaries, in-situ statistics, parallel HDF5 output and AMD GPUs, and switches cfl to the conventional unstructured normalization. Changes that affect results - Conventional CFL normalization ([#175](https://github.com/Matthe
source: github-release - Mallard 0.3.0
v0.3.0releaseMallard 0.3.0 adds three-dimensional simulation, periodic boundaries and a distributed-memory (MPI) solver that scales across GPU nodes. 3D - -DMallard_DIM=3 builds a 3D solver for tetrahedra, hexahedra, prisms and pyramids in any mix ([#35](https://github.com/MatthewBonanni/mall
source: github-release - Mallard 0.4.0
v0.4.0releaseMallard 0.4.0 completes the distributed-memory solver and adds finite-rate chemistry: reacting Euler and Navier–Stokes flows with arbitrary Cantera mechanisms. Finite-rate chemistry Following the design in docs/design/chemistry.md ([#52](https://github.com/MatthewBonanni/mallard/
source: github-release - Mallard 0.5.0
v0.5.0releaseMallard 0.5.0 makes finite-rate chemistry fast on GPUs and makes TENO-E robust on stretched cells. Chemistry on GPUs - Team-per-cell kernels ([#119](https://github.com/MatthewBonanni/mallard/pull/119)): from 16 species, each cell is integrated by a warp, with a sparse LU from 30
source: github-release - Mallard 0.2.0
v0.2.0releaseMallard 0.2.0 is the first release built with agentic development. It turns the 0.1 prototype into a validated, high-order 2D solver that runs on CPUs and NVIDIA GPUs. Documentation and gallery: [https://matthewbonanni.github.io/mallard/](https://matthewbonanni.github.io/mallard/
source: github-release