AFiD code
AFiD is an open-source Navier–Stokes solver for wall-bounded turbulence. With the international solver team, I co-developed AFiD-GPU for large parallel simulations of canonical turbulent flows.
The collaboration includes the University of Twente, University of Rome “Tor Vergata”, SURFsara and NVIDIA. The solver family supports canonical convection and shear flows; consult each repository for supported configurations.
Richard’s contribution and published scaling
The GPU methods paper credits all twelve co-authors, including Richard Stevens. The GPU implementation builds on the wider team’s AFiD solver; the CPU methods paper has a different author list.
The published GPU study reports scaling tests up to 4,096 GPU nodes on Piz Daint. These are historical, grid- and configuration-specific measurements using NVIDIA K20X/P100 hardware, not a performance guarantee for present systems. See the author manuscript and benchmark tables.
Technical and reproducibility overview
Numerical method
AFiD solves the incompressible Navier–Stokes equations, with an additional temperature or scalar equation where required, in Cartesian or cylindrical coordinates. The published CPU method uses conservative second-order centred finite differences on a staggered grid, MPI, and a two-dimensional pencil domain decomposition.
Builds and configurations
The CPU repository documents MPI, BLAS, LAPACK, FFTW3, and parallel HDF5 prerequisites and provides a manual and example inputs. The GPU implementation uses CUDA Fortran and supplies CPU, GPU, and experimental hybrid builds. Its machine-specific instructions are historical and must be adapted and tested on current systems.
Versioning, licence, and verification
The latest tagged CPU release is v1.1 (January 2016), while the main branch contains later commits. Reproducible studies should therefore record the exact commit, compiler and libraries, input file, grid, and case-specific convergence checks. The GPU repository is MIT licensed; the CPU repository contains separate copying terms that should be read before reuse.
Using or extending AFiD. For a scientific collaboration, email Richard Stevens with the flow configuration, parameter regime, target observables, intended hardware, and the repository commit you have tested. The public repositories are research code; the cited papers and case-specific checks, rather than repository availability alone, establish numerical evidence.
When using the GPU version, please cite:
X. Zhu, E. Phillips, V. Spandan, J. Donners, G. Ruetsch, J. Romero, R. Ostilla-Mónico, Y. Yang, D. Lohse, R. Verzicco, M. Fatica, R.J.A.M. Stevens, AFiD-GPU: A versatile Navier–Stokes solver for wall-bounded turbulent flows on GPU clusters, Comput. Phys. Commun. 229, 199-210 (2018).
When using the CPU version, please cite:
E.P. van der Poel, R. Ostilla-Mónico, J. Donners, R. Verzicco, A pencil distributed finite difference code for strongly turbulent wall-bounded flows, Computers & Fluids 116, 10-16 (2015).
The images below were generated from simulations performed with AFiD.
Rayleigh-Bénard convection
For corresponding video look at the Physics of Fluids YouTube channel. Further information is available on the thermal convection research page.
Visualization by E. van der Poel and R. Ostilla-Mónico.
Sheared convection
For corresponding video look at the Physics of Fluids YouTube channel.
From: A. Blass, X. Zhu, R. Verzicco, D. Lohse, R.J.A.M. Stevens - Direct numerical simulations of sheared thermal convection, Winner of the 2017 SURFsara Visualization Competition.
Double diffusive convection
For corresponding video look at the Physics of Fluids YouTube channel.
From Y. Yang, E. van der Poel, R. Ostilla-Mónico, C. Sun, R. Verzicco, S. Grossmann, and D. Lohse, Salt fingers in double-diffusive convection bounded by two parallel plates. Submitted to the APS Division of Fluid Dynamics Gallery of Fluid Motion in 2014.
Taylor-Couette flow
Visualisation by: X. Zhu, radius ratio η=0.714 for Ta=109.