The parallel algorithm for the gravity structural direct problem solution on the GPU

One of the main limiting factors for geophysical modelling today is the grid size. It is very limited with available computer resources. In this paper author describes the graphical processing units (GPU) usage for geophysical modelling purposes. Gravity direct structural problem is considered which is an example of algorithm of fourth complexity degree (O(N×M×N×M), where N×M is a grid size). Idea of GPU parallel algorithm for this problem solution is described and the algorithm block scheme is presented. Also the simple calculation for the grid of 256×256 size was performed on different hardware and results were compared. Even the common home-class graphic card demonstrated the 64× speed gain in comparison with a single CPU calculation. The usage of 8 science-class Tesla GPUs in parallel showed the 752× times bigger calculation speed than the single CPU. Thus, the GPU usage has a great potential in the geophysics and gives one a possibility to construct portable and cheap high-efficient supercomputer which performance could be compared with previous generation of “classical” supercomputers performance.