GPU-Based Parallel Reservoir Simulators

摘要: Large-scale reservoir simulation demands significant computational time so improving its efficiency becomes crucial. Graphics Processing Unit (GPU), a high-profile parallel processor with hundreds of microprocessors, offers great potential in because efficient power utilization and high efficiency. In addition, cost is relatively low, making large-scale possible for most desktop users. this presentation several GPU-based linear solvers preconditioners will be discussed. They include the GMRES, BiCGSTAB ORTHOMIN incomplete LU (ILU) factorization, domain decomposition algebraic multigrid preconditioners. These have been coupled an in-house black-oil simulator to speedup simulation. numerical experiments performed, SPE 10 problem, 3D heterogeneous benchmark model over one million grid blocks, selected test resulting simulator. On state-of-the-art CPU GPU platforms, new implementation can achieve eight times solving systems arising from problem compared implementation.