A GPU–CUDA based direct simulation Monte Carlo algorithm for real gas flows

• A cell-based GPU DSMC algorithm is proposed.
• The algorithm incorporates detailed physical and kinetic models.
• It is validated for both non-reacting and reacting flows.
• Performance is comparable to an alternate GPU code despite more advanced physics.
• Other benefits of the cell based strategy are discussed.

Here we describe a Direct Simulation Monte Carlo (DSMC) algorithm that takes advantage of the computational performance of Graphics Processing Units (GPUs) to simulate rarefied flows where real gas effects such as internal relaxation and chemical reactions may be present. The algorithm, currently written in 2D, uses adaptive mesh generation to limit the number of particles in each cell for maximum computational efficiency, and collision sub-cells for increased accuracy. Results are presented for the Mach 10 flow of argon over a circular cylinder at Kn = 0.01, the flow of chemically reacting nitrogen over a 62.5° wedge at Kn = 0.0005, and the supersonic flow of argon over a block at 30° angle of attack. The code achieves a performance of 20 ns/simulator particle/time-step (non-reacting flow) and 30 ns/simulator particle/time-step (reacting flow) on a single 448 core GPU and is approximately 20 times faster than the DS2V code running on a single CPU. This makes the DSMC method more applicable by partially alleviating the main limitation of long computational run-times.