Accelerating 3D ocean model development by using GPU computing

Three-dimensional numerical ocean models should cost large amount of computational time. To save computational time, most ocean models choose to deduce the amount of computation, by using mode splitting technique or implicit technique, while these techniques might influence accuracy. We are now developing a new three-dimensional ocean model without using computation-reduction techniques, which directly and explicitly solves the governing equations of ocean circulation. To accelerate the model's running, PGI Accelerator Visual Fortran is used to code the model program. This Fortran language software enables our model to use CUDA, a parallel computing architecture from NVIDIA, which brings dramatic increases in computing performance by harnessing the power of the GPU (graphics processing unit). Our accelerated model runs 20 times faster on a NVidia GTX460 GPU than on an Intel Core i7-920 CPU, while the model's results remains almost the same. Thus, our model development becomes efficient.