The Lattice Boltzmann Method has been demonstrated to have an outstanding numerical performance especially for multi-scale, multi-component flows and flows in porous media. Due to its highly explicit nature of solution, LBM is proved to be a unique candidate for parallelism. Recently many efforts have been made to develop parallel LBM simulations on manycore Graphics Processing Units (GPU). Recent researches on implementing LBM simulations on GPUs have been very successful, showing impressive speedups even on single devices. However, the resulting computational performance strongly depends on the developers’ proficiency to harness the maximum efficiency of these new computing architectures.One of the very interesting areas to be investigated by the Lattice Boltzmann method is to employ it for simulating gas mixture flows through packed beds. Packed beds have noticeable significance in industrial systems. The complex process of binary mixture diffusion has itself several applications in fluidic systems such as chemically reacting flows, gas purification, pollutant dispersion and so forth. Diffusion of oxygen and nitrogen, discussed in the present work, is of a crucial importance since these two elements are the major constituents of atmospheric air and their diffusion plays a great role in nitrogen/oxygen purification processes. Of paramount importance for industry, is the flow of oxygen and nitrogen through highly packed beds of spherical particles with adsorption properties. In a recent research by Rastegari and Ashrafizaadeh (2009) LBM was employed to simulate the flow of oxygen and nitrogen mixture in a simple packed bed of spherical particles. Although the results are very promising from the physical point of view, the simulations suffer from extremely long computational times, leading to applying some simplifications to keep the simulation time within affordable bounds. In this thesis a 3D lattice Boltzmann flow solver for the binary flow of oxygen and nitrogen mixture has been implemented on GPUs. It is highly distinguished from the previous researches since it employs complex, highly dense packed beds where the no slip boundary condition is treated accurately on the solid surface of the spherical particles which even doubles the computational cost of the simulations in such media. As such, and to exploit the maximum computational power of GPUs an optimized algorithm is used and the flow is scaled in a way that the traort equations for both species can be managed in parallel. It is shown that, using a wide range of modern graphics processors, either single or in parallel, speedups of more than two orders of magnitudes over single CPUs are achievable; bringing the long, expensive simulations down to reasonable costs and times. Keywords: Graphics processing, Parallel processing, Lattice Boltzmann method, Packed beds, Gas mixtures.