An agent-particle on GPU for deterministic convection-diffusion

摘要

This paper proposes an implementation of wellknown convection-diffusion processes that is realized entirely on GPU with CUDA. Indeed, these dynamics are widely observed in biological areas such in displacements of millions of cells specially during aggregations or repulsions movements. This implementation rests on mesh-free method instead of gridbased method. Therefore we used the agent-particle formalism described here and which is based on SPH (Smoothed Particle Hydrodynamics) numerical method and agent-based modelling. To perform convection-diffusion, neighboring search, density computation and velocity and position update are the main stages of computing. While agent level simulations require high degree of computational power to show realistic behaviors, we use the GPU computing to drive large-scale simulation. The results show more increased performance compared to CPU and suggest that GPU is a valuable tool to deal with complex systems simulation such in biology.