GREEN SUPERCOMPUTING � CALCULATING TSUNAMI WAVE PROPAGATION

Nowadays, the use of supercomputing for solving almost any task ranging from industrial to the ecological ones has been increasing. It should be noted that this requires lots of energy, which negatively affects the environment and is costly. However, in a number of cases it is possible to achieve high performance calculations without the use of supercomputers but instead by relying on special processors, focused on a particular class of tasks. While optimizing the computational pipeline for a particular mathematical model, the modern Field Programmable Gates Array (FPGA) platform provides an opportunity to accelerate the numerical solution of a problem by orders of magnitude using a personal computer (PC). This point is here discussed for the case of tsunami modeling. Underwater earthquakes can cause destructive tsunami waves resulting in significant casualties and economic losses. Therefore, after each seismic event, the hypocenter of which is underwater, many supercomputer centers start the modeling process in order to determine the coastal areas where the tsunami wave may cause a loss human life or infrastructure destruction. This certainly requires a lot of energy. The paper shows the possibility of numerical modeling of transoceanic tsunami wave propagation on a grid with a step of 1 geographic minute using a PC platform, which demonstrates significant performance compared to supercomputer calculations. The modeling process (numerical solution of the system of differential equations of shallow water, well describing the process of tsunami wave propagation) takes only a few minutes using a PC with hardware accelerator. Such performance, comparable to the capabilities of a modern computing cluster, is achieved thanks to hardware acceleration - a specialized FPGA-based calculator used as a co-processor. The correctness of the solution is confirmed by comparison with known exact solutions of the system of shallow water equations. The limitation of the proposed method is the size of the random access memory of the device, since high performance is achieved provided that there is no necessity to use the hard disc memory of the computer. The application of the proposed computation technology will allow use �green supercomputing� in many cases.