Simulation of droplet dynamic with high density ratio two-phase lattice Boltzmann model on multi-GPU cluster

Three-dimensional liquid droplet dynamic is simulated by a high- density ratio two-phase lattice Boltzmann model of Lee and Liu [1] on multi-GPU cluster. The procedure is first used to compute phase separation of the randomly mixed gas-liquid mixture. The capability of the model is further validated by contrasting the predicted oscillation frequency of the perturbed droplet and merging droplets with analytical solutions. For multi-GPU implementation, overlapping computation and communication is shown to improve up to 50% of performance. Strong scaling test is adopted to examine the scalability of the multi-GPU implementations. The performance of nVIDIA Tesla P100, GTX Titan, GTX Titan Z are respectively 10, 2.8 and 2.6 times faster than nVIDIA Tesla M2070. The present implementation also shows that 12 GTX Titan GPUs is approximately 190 times faster than 12 cores of CPU (i7-4930K).