high performance computing on graphics processing units: hgpu.org

In this article a strategy to accelerate the simulation and visualization of combined advective-diffusive mixing of a contaminant inside a square cavity with time-dependent boundary-conditions is presented. No moving walls are required to mix the fluid, but natural convection by periodic temperatures on opposite walls. A contaminant will diffuse asymptotically to uniform concentration. Advective mixing is evaluated with Lagrangian tracking and diffusive mixing is calculated with the Diffusive Strip method. That calculation requires hours of CPU time due to the large amount of numerical operations and the precision requested but it can be directly translated to a Graphics Processors Unit (GPU), improving the performance by orders of magnitude. The algorithms implemented in Compute Unified Device Architecture (CUDA) and an analysis of the speed achieved are presented. The visualization of the diffusion process is done simultaneously using the data stored in the GPU memory, which allows to make a real-time analysis of the mixing.