Research Of The Lattice Boltzmann Method Based On GPU Parallel Computing

Lattice Boltzmann method (LBM) is a novel numerical simulation methodthat developed in recent decades. The advantages of the boundary conditions thatare easy to handle, good calculation stability, the natural data parallelism play animportant role in computational fluid dynamics.However, with the simulation scale and the amount of computingincreasing, to improve computing performance, only increasing the number ofCPU’s nuclear is very limited. Using the traditional multi-core parallelcomputing method can not complete the very large scale computing tasks. So oneconsiders that the graphics processing unit GPU is applied to numericalcomputing. Traditionally, the GPU is only responsible for graphics rendering.Most of the work are completed by the central processing unit CPU. GPU andCPU has a different design idea. The GPU has a unique advantage in parallelcomputing. Therefore, GPU-based parallel computing will have a broad prospectfor large-scale computation simulation tasks. Taking into account the latticeBoltzmann method based on GPU high performance computing provide a newidea for people.The paper firstly computed matrix multiplication of different order basedon GPU parallel computing. The experimental results show that the greater theorder, the more significant the speedup of parallel computing and serialcomputation is. It fully reveals the superiority of GPU on the computing rate.Moreover, the experimental results verify that the use of different types ofstorage affect the calculation speed. The use of shared memory can furtherimprove the calculation speed. Then, we design a LBM algorithm oftwo-dimensional single-component two-phase flow based on GPU parallelcomputing. The boundary condition is periodic boundary condition. The model isD2Q9. On the basis of the two-dimensional algorithm, we design a LBMalgorithm of three-dimensional single-component two-phase flow based on GPUparallel computing. Using D3Q19model and periodic boundary condition. Then,based on the above parallel algorithm, we simulate the process of the phaseseparation of two-dimensional and three-dimensional single-component two-phase flow in the GPU high performance computing platform. Theexperimental results show that the GPU parallel computing is about10.2timesfaster than the CPU serial computing platform, we simulate the cavitationphenomenon which is common in people’s lives. The experiments showed thewhole process of the tiny air nuclear in water taking shape, swelling andcollapsing. The experimental results equally show the advantage of GPU parallelcomputing speed.