Enhancing regional ocean modeling simulation performance with the Xeon Phi architecture

Abstract

Ocean studies are crucial to many scientifc disciplines. Due to the diffculty in probing the deep layers of the ocean and the scarcity of data in some of the oceans, the scientifc community relies heavily on ocean simulation models. Ocean modeling is complex and computationally intensive, and improving the performance of these models will greatly advance and improve the work of ocean scientists. This paper presents a detailed exploration of the acceleration of the Regional Ocean Model System (ROMS) software with the latest Intel Xeon Phi x200 architectures. Both shared-memory and distributed-memory parallel computing models are evaluated. Results show run time improvements of nearly a factor of 16 compared to a serial implementation. Further experiments and optimizations, including the use of a GPU acceleration model, are discussed and results are presented.

Parallel Implementation and Optimization of Regional Ocean Modeling System (ROMS) Based on Sunway SW26010 Many-Core Processor

“…There are many excellent optimizations of ocean numerical model programs based on various kinds of CPUs and accelerators. Lupo et al [22] used OpenACC [11] to port ROMS on Intel Xeon Phi X200 [23], and achieved good acceleration. Yalavarthi and Kaginalkar [24] ported ROMS on Intel MIC [13], and model 2D kernel is executed as the hotspot on the co-processor.…”

A Scalable Semi‐Implicit Barotropic Mode Solver for the MPAS‐Ocean

“…As stated in several previous studies, solving the barotropic mode is a major bottleneck in ocean models (Hu et al, 2015;Huang et al, 2016;Koldunov et al, 2019;Lupo et al, 2017;Prims et al, 2019). There have been two dominant approaches to solve the barotropic mode, explicit-subcycling and semi-implicit.…”