Parallelizing Broad Phase Collision Detection for Animation in Games: A Performance Comparison of CPU and GPU Algorithms

Games, computer animations and three-dimensional interactive simulations have required the use of realistic and faster than ever before broad phase collision detection algorithms. In this work, we compare the performance of four broad phase algorithms implemented on CPU and GPU, using four different test scenarios. More specifically, one of them is a new GPU-based algorithm that we have developed in the Bullet library using CUDA, and the other three remaining implementations are CPU-based algorithms available in the same library. The experimental results show that the heterogeneous algorithm is competitive when compared to some robust methods available in Bullet, particularly in scenes with a large number of objects whose movements are complex and unpredictable. We believe that initiatives like this, which explore solutions for new implementations of collision algorithms running on GPU and operating asynchronously with the CPU, are extremely important and useful for game designers, especially in the area of digital games based on Physics, considering there are other elements of the animation, e.g., sound and artificial intelligence, which can thus be executed during the broad phase calculation.