GauRast: Enhancing GPU Triangle Rasterizers to Accelerate 3D Gaussian Splatting

Abstract--3D intelligence leverages rich 3D features and stands as a promising frontier in AI, with 3D rendering fundamental to many downstream applications. Previous efforts to accelerate 3DGS rely on dedicated accelerators that require substantial integration overhead and hardware costs. These platforms are increasingly crucial due to AI by leveraging rich 3D features to enhance understanding the growing demand for 3D processing in mobile and embedded and interaction within complex environments. Specifically, 3DGS achieves only Fei Li, co-founder of ImageNet, emphasized, "...we need 2-5 FPS on these platforms [22] with commonly used realworld, spatially intelligent AI that can model the world and reason large-scale datasets [3], falling short of the performance about objects, places, and interactions in 3D space and requirement for most practical applications. This underscores the importance of 3D intelligent gap poses challenges for deploying advanced 3D intelligence applications such as autonomous driving [39], robotics [32], in resource-constrained environments, highlighting the need and augmented/virtual reality (AR/VR) [4] shown in Figure 1.