The Research And Implementation Of CUDA-Based Parallel 9*9 Go Engine

In Computer Game, Computer Go is the most challenging one. The goal of Computer Go is to simulate multiple possible move sequence from any given board to the end of the game. The most optimized move for the given board could be calculated by the results of the simulations. An accurate evaluation for a board must be achieved by searching. Time limit is an important restriction to Computer Game. The Computer Game performance is directly related to the accuracy of board evaluation and the number of boards evaluated in given time. Because the performance of each core is restricted by material, power consumption and heat, modern processors harvest performance increase by increasing the number of cores. The multi-core/many-core processor is a challenge to the conventional single-threaded software. The research on parallelized Computer Go algorithm for the modern processor is promising to enhance the Computer Go performance.This article summarize the development of Computer Go, proposed a GPU-based Go engine implementation, improved the Monte-Carlo Tree search algorithm suitable for CUD A GPU hardware implementation, realized the CUDA-based parallel Go engine. The test result shows the Go engine could achieve a simulation for 50,000 board send, which is 5.0 times of a single-core CPU. CUDA is usually deployed for massive parallel computing. In the fields other than numerical computing, CUDA is rarely used. Applying CUDA in Go engine introduced in this article is a breakout. In the implementation, bit-operation is applied for higher performance.