PhotoNs-GPU: A GPU accelerated cosmological simulation code

Abstract We present a GPU-accelerated cosmological simulation code, PhotoNs-GPU, based on an algorithm of Particle Mesh Fast Multipole Method (PM-FMM), and focus the GPU utilization optimization. A proper interpolated method for truncated gravity is introduced to speed up special functions in kernels. verify code mixed precision different levels theinterpolated GPU. run with single roughly two times faster than double current practical simulations. But it could induce unbiased small noise power spectrum. Compared CPU version PhotoNs Gadget-2, efficiency new significantly improved. Activated all optimizations memory access, kernel concurrency management, peak performance our test runs achieves 48% theoretical average approaches ∼35%