Parallelized CCHE2D flow model with CUDA Fortran on Graphics Processing Units.

Highlights: [•] The Parallel Cyclic Reduction (PCR) method was used to parallelize the ADI solver. [•] The parallelized SIP solver using Red–Black Checkerboard method was not stable. [•] Allowing multiple equations solved at one thread resolved size limitation problem. [•] CCHE2D implicit flow model was successfully parallelized using CUDA Fortran. [•] Much higher computing efficiency and comparable accuracy was obtained on GPU. [ABSTRACT FROM AUTHOR]