Game Development in CUDA

Abstract

We explore the feasibility of running an entire game on the GPU using CUDA, aiming to harness parallelprocessing for large-scale simulations. By offloading all game logic and mechanics to the GPU, wesuccessfully managed to make a game containing a million of dynamic entities, showcasing the possibilityof supporting far more than typical CPU-based designs. The project takes the form of a city simulation,serving as a proof-of-concept for a fully GPU-driven approach to game development.Developing this game in CUDA means building core systems from scratch and adapting traditional gamemechanics to fit the GPU’s parallel structure. Since current engines rely on CPU-GPU collaboration,working without a GPU-focused engine presents unique challenges but also emphasizes the potential ofa GPU-only approach.Our results indicate that GPU-driven games could make way for new gameplay mechanics and richersimulations by handling complex, high-entity interactions and large environments more efficiently. Webelieve that this method could become a viable option for developers aiming to push the boundaries oflarge-scale, real-time simulations.