Accelerated Many‐Core GPU Computing for Physics and Astrophysics on Three Continents

Abstract

PrefaceTo be completed.Complex systems modeling and simulation approaches are being adopted in a growing number of sectors, including finance, economics, biology, astronomy, and many more. Technologies ranging from distributed computing to specialized hardware are explored and developed to address the computational requirements arising in complex systems simulations. The aim of the book is to present a representative overview of contemporary large-scale computing technologies in the context of complex systems simulation applications. The intention is to present the state of the art and to identify new research directions in this field and to provide a communication platform facilitating an exchange of concepts, ideas and needs between computer scientists and technologists and complex systems modelers. On the application side, the book focuses on modeling and simulation of natural and man-made complex systems, because high-level requirements are similar across a wide range of domains. On the computing technology side emphasis is placed on distributed computing approaches, but supercomputing and other novel technologies (e.g., specialized hardware) are also considered. Coleraine,Werner Dubitzky

Directionally unsplit hydrodynamic schemes with hybrid MPI/OpenMP/GPU parallelization in AMR

“…This issue was found to be of minor importance in the simulations with higher spatial resolution (and, hence, lower surface/volume ratio). For example, for the 128-GPU benchmark on the Laohu GPU cluster at NAOC, the MPI communication time takes less than 2% of the total execution time (Spurzem et al 2011). Also note that this issue can potentially be largely alleviated by overlapping communication with computation.…”

“…In astrophysical simulations, considerable performance speed-ups in multi-GPU systems have been demonstrated in a broad range of applications, for example, the direct N -body simulations (e.g. Schive et al 2008; Gaburov et al 2009; Spurzem et al 2011), Barnes–Hut tree algorithm (Hamada et al 2009), reionization simulations (Aubert and Teyssier 2010), adaptive mesh refinement (AMR; Schive et al 2010; Wang et al 2010), general relativistic magnetohydrodynamics (MHD) (Zink 2011), and interactive visualization of large 3D astronomical data (Hassan et al 2011).…”

“…A hybrid CPU/GPU model is adopted in the code, in which both hydrodynamic and gravity solvers are implemented into GPU and the AMR data structure is manipulated by CPU. An order of magnitude performance speed-up is demonstrated on the Laohu GPU cluster at the High Performance Computing Center at National Astronomical Observatories of China (NAOC), using up to 128 Tesla C1060 GPUs (Spurzem et al 2011). The code has been applied to different aspects of astrophysics (Shukla et al 2011).…”

Expansion techniques for collisionless stellar dynamical simulations

“…They are used to accelerate many scientific calculations, especially in astrophysics, such as the dynamics of dense star clusters and galaxy centers (see review by Spurzem et al 2012). The main issue when implementing a mathematical technique for GPUs is to make the algorithm parallel.…”

Dynamical model of an obscuring clumpy torus in AGNs – I. Velocity and velocity dispersion maps for interpretation of ALMA observations

“…The code is designed to use different softening parameters for the gravity calculation (if it is required) for different astrophysical components in our simulations like SMBHs, dark matters or stars particles. More details about the ϕ−GPU code public version and its performance are presented in (Spurzem et al 2012;Berczik et al 2013). The present code is well tested and has already been used to obtain important results in our earlier large scale simulations (up to few million bodies) (Wang et al 2014;Zhong et al 2014;Li et al 2012Li et al , 2017.…”