Parallel alpha-beta algorithm on the GPU

Abstract

In the paper we present the parallel implementation of the alpha-beta algorithm running on the graphics processing unit (GPU). We compare the speed of the parallel player with the standard serial one using the game of reversi with boards of different sizes. We show that for small boards the level of available parallelism is insufficient for efficient GPU utilization, but for larger boards substantial speed-ups can be achieved on the GPU. The results indicate that the GPU-based alpha-beta implementation would be advantageous for similar games of higher computational complexity (e.g. hex and go) in their standard form.