High-Performance Computation of LGCA Fluid Dynamics on an FPGA-Based Platform

Lattice Gas Cellular Automata (LGCA) simulations are typical High-Performance Computing (HPC) applications commonly used to simulate fluid flows. Due to the computational locality and discretization of LGCA, these simulations can achieve high performance by using parallel computing devices like GPUs or multi-core CPUs. Nevertheless, many studies also have shown that state-of-the-art Field Programmable Gate Arrays (FPGAs) have enormous parallel computing potential and power-efficient for high-performance computations. In this paper, we present an FPGA-based fluid simulation architecture design for the LGCA method. Our design exploits both temporal and spatial parallelism inside the LGCA algorithm to scale up the performance on FPGA. We also propose an application-specific cache structure to overcome the memory bandwidth bottleneck. Furthermore, our development process is based on the High-Level Synthesis (HLS) approach that increases productivity. Experimental results on a Xilinx Vcu 1525 FPGA show that our design is able to achieve 17130.2 Million Lattice Updates Per Second (MLUPS).