CUDA at Scale Independent Project
This project presents a GPU-based implementation of the classic Tic-Tac-Toe game where two automated players compete against each other using CUDA kernels. Instead of performing move calculations on the CPU, the game uses parallel GPU execution to analyze the board and decide moves efficiently. Each player follows a different strategy to make the gameplay more dynamic and demonstrate multiple approaches to move selection on the GPU.The first GPU player uses a random-move strategy, where CUDA threads search for available positions and select a valid move. The second GPU player uses a structured strategy that scans the board and selects the first available empty position using parallel evaluation and synchronization techniques. CUDA atomic operations are used to safely coordinate thread access while choosing moves.The game board is updated after every turn, and the current game state is displayed continuously until a player wins or the match ends in a draw. The project demonstrates important CUDA concepts such as kernel execution, parallel thread computation, GPU memory handling, synchronization, and parallel decision-making.The main objective of the project is to show how GPU computing can be applied not only to scientific computations but also to game logic and AI-based decision systems. By allowing two GPU-controlled competitors to play against each other, the project highlights the ability of CUDA to process multiple possibilities simultaneously and improve computational efficiency in interactive applications.