Motivation and Design Principles

Sailfish is a general purpose fluid dynamics solver optimized for modern multi-core processors, especially Graphics Processing Units (GPUs). The solver is based on the Lattice Boltzmann Method, which is conceptually quite simple to understand and which scales very well with increasing computational resources.

The Sailfish project is also an experiment in scientific computing and software engineering. Unlike the majority of CFD packages, which are written in compiled languages such as C++ or Fortran, Sailfish is implemented in Python and CUDA C/OpenCL. We have found this combination to be a very powerful one, making it possible to significantly shorten development time without sacrificing any computational performance.

The general goals of the project are as follows:

• Performance: the code is optimized for the current generation of NVIDIA GPUs. With a single state-of-the-art video board, it is possible to achieve a simulation speed of 800 - 1200 MLUPS (depending on the used grid and model). To achieve comparable performance with typical off-the-shelf CPUs, a small cluster would be necessary.
• Scalability: the code is designed to scale well with increasing number of compute cores.
• Agility and extendability: by implementing large parts of the code in a very expressive language (Python), we aim at encouraging rapid experimentation. Running tests, playing with new boundary conditions or new models is easy, and it often only requires changing a few lines of the kernel code.
• Maintainability: we keep the code clean and easy to understand. The Mako template engine makes it possible to dynamically generate optimized code without any unnecessary cruft.
• Ease of use: defining new simulations and exploring simulation results is simple and many details are automated and by default hidden from the end-user.