The Application Of High-Performance GPU Technology In The Lattice Boltzmann Method Simulation

Based on the existing theories and models as well as the computing power, Computer simulation utilizes the Numerical simulation methods to take the virtual experiments. The efficient and accurate computer simulation needs for models, computational methods, software and hardware common to improve and close coordination. Fluid Mechanics is a subject which studys of fluid motion and the law, the research is very extensive, even after years of development and exploration has yielded fruitful results, but the field still faces enormous challenges:on the one hand, is due to fluid motion complexity; the other hand, As modern science and technology fast development, fluid continuously with other disciplines, generated a lot of hard to describe traditional methods or new problems do not apply. The Lattice Boltzmann method (Lattice Boltzmann Method, LBM) in recent decades the international community developed a new method of computational fluid dynamics, it is from lattice gas automata (Lattice Gas Automata, LGA) from the technical development. LBM as an alternative method of Computational Fluid Dynamics Simulation of people have been recognized and widely used in fluid systems with complex geometry boundary simulation problems. LBM inherits the LGA completely parallel, the boundary conditions easy to implement, easy to program, etc. In addition, while gone, such as statistical noise, the pressure depends on the fluid velocity does not satisfy the Galilean invariance and other defects. LBM unlike the traditional numerical methods is the Navier-Stokes equations at the macro, based on the finite element discretization for the poor into, we get algebraic equations, and then use a numerical method for solving the discrete equations to obtain macroscopic quantities, it is built micro-particle distribution function in the dynamic equation based on the amount of the macro-particle distribution function. Because of these advantages, the LBM is widely used in simulation of complex fluid systems, such as suspended particulates, viscoelastic media, multiphase flow, particle settling and suspended particles in a fluid motion, blood flow,etc., LBM have from the purely theoretical into the practical engineering application areas, the formation of an international research focus. In the second chapter of this paper a detailed description of the LBM, including a brief description of LGA theory, LBM's basic theories and models, all kinds of complicated boundary conditions and procedure structure. Poiseuille flow was studied by simulating a variety of curves LBM boundary conditions in the implementation and efficiency, taking into account the overall macro no-slip boundary amendment is proposed under the premise of a new unified curve boundary conditions and boundary conditions with the existing experimental comparing simulated and the error obtained in the simple boundary conditions, form a unified, in addition to the relaxation time of the singularity, and has better accuracy and so on.Although the LBM for fluid simulation is a good method, but it is a typical simulation of compute-intensive, the computing power and resources are limited. The development of the current CPU bottleneck has been encountered, both in the core architecture of the efficiency or the number of ways in the core is difficult to obtain increased dramatically. Graphics processing unit (Graphics Processing Unit,GPU), NVIDIA company's proposed high-performance image processor based parallel computing technology, which was originally intended to solve complex and onerous tasks of image processing and high image processing performance computing has achieved some success. GPU is suited for parallel computing that can be expressed as the problem of parallel computing arithmetic density is very high, and floating-point computing power much higher than the CPU, the same amount of GPU computing hardware cost significantly lower than the CPU cost of hardware. Since 1993, GPU performance to an annual growth rate of 2.8 times, basically every six months even the new generation. The rapid development of GPU graphics hardware, the emergence of a number of programming model, which is the most representative introduced NVIDIA unified architecture computing devices (Compute Unified Device Architecture,CUDA), to achieve the GPU general computing (General Purpose GPU,GPGPU), efficient solution for the problem of fluid dynamics calculations provide a new way. How to use hardware programmability and parallelism to achieve some complex calculations, has become one of the hot. The rapid development of general-purpose GPU computing, especially the "CPU+GPU" heterogeneous model proposed, the GPU is more extensive application areas:scientific computing (such as fast Fourier transform, scientific visualization, Accelerating genetic algorithm, database operations,etc.), image processing, signal processing, video and audio processing, computational biology, physics-based simulation. Which, GPU fluid system in a complex simulation application is in its infancy, research GPU technology in the LBM simulation for the deepening and the promotion of LBM in the complex behavior of the system simulation is of great significance.This paper describes the development process of GPU technology, features and advantages of modern GPU, GPU for general computing principles, and the current most widely used programming environment CUDA technology, while the GPU parallel computing system implementation of the direction, the use of LBM, respectively, CPU and GPU in the context of the Poiseuille flow and single-cylinder two-dimensional movement of fluid between channels were simulated. The results showed:Poiseuille flow in the use of single-precision type simulation, under the Debug CPU simulation time required is 26-35 times the GPU simulation, simulation of the Release under the CPU time is required for GPU simulation 4.5-6 times; Single Cylindrical two-dimensional movement of fluid between channels in the Intel (R) Core (TM) i5 CPU 760@2.80GHz CPU simulation time is under the GPU simulation 10.3 times in the Intel (R) Pentium (R) 4 CPU 3.00 GHz CPU under simulation when the simulation time is 32.3 times the GPU. GPU and CPU simulation time in the simulation, there are some errors, this paper analyzed and discussed. In short, fully illustrated by the large number of experiments GPU technology in LBM in simulation of complex fluid behavior of the feasibility and efficiency, as the GPU effectively applied to high-performance technology with complex geometric boundaries of the fluid simulation has laid a good foundation.