4 - 6 milliards d'opérations logiques par seconde : simulation de phénomènes naturels par automates cellulaires en utilisant une matrice SIMD bidimensionnelle (GAPP). Application à la diffusion de particules

The « cellular automata » approch (see the references [8] to [17]) to the simulation of natural phenomenon leads up to computation problems essentially due ta the required computation power and to the data transfer bottleneck between processing elements and memories . A real-time image processing system for low and medium level processing that we developped at LSIT, Strasbourg, enabled us to solve these problems. A systolic SIMD array (GAPP of NCR) gaved us enough processing power to handle the iterative cellular automata method . In this paper, we will first describe our image processing system (MASYVE) and give the simulated diffusion model . We will then describe the management of data transfers . At last we will present the achieved performances, the limitations of the system and the perspectives .