10M-Core Scalable Fully-Implicit Solver for Nonhydrostatic Atmospheric Dynamics
SessionACM Gordon Bell Finalist II
Session ChairSubhash Saini
Event Type
ACM Gordon Bell Finalist
Accelerators
Algorithms
Effective Application of HPC
Performance
Scientific Computing
TimeThursday, November 17th11:30am - 12pm
Location255-EF
DescriptionAn ultra-scalable fully-implicit solver is developed for stiff time-dependent problems arising from the hyperbolic conservation laws in nonhydrostatic atmospheric dynamics. In the solver, we propose a highly efficient hybrid multigrid domain decomposition preconditioner that can greatly accelerate the convergence of the solver at the extreme scale. For solving the overlapped subdomain problems, a physics-based multi-block asynchronized incomplete LU factorization method is customized to further exploit the on-chip fine-grained concurrency. We perform systematic optimizations on different hardware levels to achieve best utilization of the heterogeneous computing units and a substantial reduction of data movement costs. The fully-implicit solver enables fast and accurate atmospheric simulations on the new Sunway TaihuLight supercomputer in China, scaling to over ten million heterogeneous cores and achieving a sustained performance of over two petaflops.
Download PDF
Paper provided by the IEEE Computer SocietyPaper also available from the ACM Digital Library
Authors
