Transactions on GIGAKU 4(1)(2017)04006/1-8 1 Parallel implementation of Entropic lattice Boltzmann method for flow past a circular cylinder at high Reynolds number Ayurzana Badarch 1,* , KhenmedekhLochin 2 , Hosoyamada Tokuzo 3 1) Graduate School of Engineering, Nagaoka University of Technology, 2) School of Applied Science, Mongolian University of Science and Technology, Mongolia 3) Department of Civil and Environmental Engineering, Nagaoka University of Technology 1603-1 Kamitomioka-machi, Nagaoka 940-2188, Japan *E-mail: ayur_426@yahoo.com Lattice Boltzmann method (LBM) has been receiving enormous attention from researcher to solve fluid related phenomena. LBM in fluid dynamics simulation, one meet some difficulties, for instance it becomes unstable for high Reynolds number flows and requires computational time and memory for large scale simulations. To avoid this, we used here the Entropic LBM (ELBM) of Karlin’s group [7] and implemented parallel code on graphical processing units (GPU). For parallel computation, we used GPGPU system of Nagaoka University of technology, which is equipped with Tesla M2050 processors. To verify accuracy and stability, we have solved double periodic shear layer flow in serial computation. The accuracy and stability of simulation by ELBM were superior to the simulation of standard LBM. Simulations of flow past a circular cylinder is carried out by parallel ELBM, where Reynolds number varies up to 140000. Using the GPU in simulation, computation time speeds up until 10 times faster than that of using central processing units (CPU). The results show that the parallel code of ELBM can solve two-dimensional turbulent flow in arbitrary geometry at the higher stability condition without using any other stabilization techniques. 1. Introduction The simulation of traditional computational fluid dynamics deals with Navier-Stokes equation (NSE). Unlike this, lattice Boltzmann method (LBM) solves the discrete Boltzmann kinetic equation in its simplified form, known as Bhatnagar-Gross-Krook (BGK) equation. Founded in the eighties of last century, LBM become recently a widely used method of computational fluid dynamics (CFD). In LBM, the fluid is simulated by molecular velocity distribution functions with discrete velocities on regular lattice. There are two operations on the distribution functions: streaming and collision. The fluid molecules propagation is described by streaming, which is non-local operation against the local collision operator, describing the collision between molecules [1]. LBM has its intrinsic feature: easy to parallelize on computer [2]. On the other hand, one of long standing problem in fluid flow simulation is the simulation of turbulent flow at high Reynolds number. In CFD, the turbulent flow simulations are usually performed either by computationally expensive direct numerical simulation of NSE or by solving averaged NSE [3, 4, 5, 6]. Direct application of the LBM for the turbulent flow simulation leads to instability of computation, because of the low viscosity of fluid. Laminar flows at low Reynolds number (Re) become turbulent when Re reached the value of Re=300 in the case of flow past a bluff body. One of the promising method of stabilizing the LBM turbulent computation is the entropic LBM (ELBM) of Karlin’s group, in which, the distribution functions are straightened to satisfy the maximum condition of the entropy at the every time step of simulation. In this paper, first we confirm accuracy and performance of the ELBM solving double periodic shear layer flow. Then we perform the calculations of high Reynolds number turbulent flow past a circular cylinder using the ELBM of Karlin et al [7, 8] in parallel algorithm.