主辦單位
ASROC
協辦單位
ASIAA
贊助單位
MOST
會徽Logo
名稱Title
2020天文年會
中研院天文所 國際會議廳

論文摘要

Small- and large-scale dynamos as the first astrophysics case for GPU API Astaroth

[ Poster ]

Miikka Väisälä (ASIAA); Johannes Pekkilä (Aalto University); Maarit Käpylä (Aalto University, Max Planck Institute for Solar System Research, Nordita); Matthias Rheinhardt (Aalto University); Hsien Shang (ASIAA); Ruben Krasnopolsky (ASIAA)

Astaroth is an application programming interface (API) developed for computing high-order 3d stencil operations, such as finite difference derivatives, using graphics processing units (GPUs). It was originally developed for performing magnetohydrodynamics (MHD) computations, although in its current state it can be adapted for any general purpose where 3d stencils are needed. The properties of the Astaroth API can be generalized for the given task via its Domain Specific Language (DSL). Astaroth is able to compute the high-order operations efficiently with methods inspired by vertex pipelining approach utilized in computer graphics, and we have benchmarked more than 30 times speedup in comparison to a reference code on a CPU node. To prepare the code for widespread astrophysical use, we explored isothermal resistive MHD turbulence, where turbulence was driven by a forcing function. We induced both small- and large scale dynamos by utilizing both helical and non-helical turbulence driving. To test for convergence we used multiple resolutions for each parameter (driving mode of turbulence and resistivity/viscosity), getting results for multiple Reynolds numbers. Our primary focus was the dynamo growth. As a function of growing Reynolds number, helical and non-helical forced systems have two separate branches of growth rates corresponding to large- and small-scale dynamos. However, the growth rates approach each other at high Reynolds numbers, which indicates that the small-scale dynamo is dominating the initial growth. Small-scale dynamo growth rates change as a function of Reynolds number in logarithmic fashion. Our results show that it is not straightforward to separate small- and large scale dynamos from each other during the dynamo growth stages. Based on the analysis of their power spectra and magnetic field distributions over time, there are signatures of a small-scale dynamo even at intermediate Reynolds numbers before a large-scale dynamo emerges.