Shallow water simulations on multiple GPUs

摘要: We present a state-of-the-art shallow water simulator running on multiple GPUs. Our implementation is based an explicit high-resolution finite volume scheme suitable for modeling dam breaks and flooding. use row domain decomposition to enable multi-GPU computations, perform traditional CUDA block within each GPU further parallelism. shows near perfect weak strong scaling, enables simulation of domains consisting up-to 235 million cells at rate over 1.2 gigacells per second using four Fermi-generation The code thoroughly benchmarked three different systems, both high-performance commodity-level systems.