al0vya/fv1_gpu

GPU-parallelised one-dimensional shallow water model

FV1-GPU

Model description

This is a successor of FV1-CPU that employs the finite volume (FV1) scheme to solve the one-dimensional shallow water equations in parallel on single NVIDIA GPUs using CUDA, called 'FV1-GPU'. Please read the information in the preceding link for a brief description of the FV1 scheme.

Running the model

FV1-GPU requires having the CUDA Toolkit installed. It has been developed using CUDA Toolkit 10.0+ and on a GPU of Compute Capability 6.1; no other Compute Capabilities have been tested. It has the same 6 test cases as FV1-CPU. After building and then running the executable, the user must select:

• Which test case to run
• How many cells are to comprise the mesh

Speedups relative to FV1-CPU

Below are speedups of FV1-GPU over FV1-CPU for a 40 s wet dam break simulation, run in release mode on a machine with an i5-8250U CPU and GTX 1050M GPU. L dictates the number of cells in the mesh as 2^L.