Dear community
We would like to announce the new open-source LB code STLBM, which is described in the pre-print Cross-platform programming model for many-core lattice Boltzmann simulations and available on GitLab.
You will find answers to the following questions:
- Is it possible to write a LB solver that runs on CPUs and on GPUs, with performances close to the state-of-the-art, using only standard C++, without libraries or language extensions ?
- Is it possible to exceed 300 MLups on a single CPU (no MPI, no GPU) ?
- Is it possible to run modern collision models (TRT, Recursive-Regularized, Cumulant, …) on GPUs with less than 50% performance loss compared to second-order BGK ?
(Spoiler alert: the answer is 3x yes…)
We are looking forward to feedback from the community and are now actively investigating how the methods of STLBM could be used to accelerate parts of Palabos on GPUs.
Cheers
Jonas, Christophe, Joël