A system design using FPGA for large scale computing

This study's objective is to define technical issues along the path to realizing an FPGA-based custom computer for CFD: computational fluid dynamics. In order to do this, we have sub-objectives, i.e. to compare calculation performance between a commodity CPU (Intel chip) and commodity FPGA (Xilinx chip), and to determine the feasibility of making a 100x faster custom computer and evaluating its performance. By doing the feasibility study, we specify some technical issues to realize a system using FPGA for large scale computiation. (1)High bandwidth data translation, (2)Circuit downsizing, (3) Performance evaluation for the CFD code as a whole, (4)Distrubited implementation over multi FPGA chips, (5)Interface with host computer, (6)Refinement of floating point processing circuit, (7)Stability, (8)Convert method in FORTRAN program to FPGA, (9)Comparison against parallel computer, (10)Implementation method/policy for high-cost/serialized functions.