Abstract: 3D Gaussian Splatting (3DGS), as an emerging explicit three-dimensional scene representation technique, has garnered significant attention in the field of neural rendering due to high image quality and efficient rendering. Unlike implicit neural methods using multilayer perceptrons, 3DGS models the scene through differentiable Gaussian distributions. It performs fast volumetric rendering without neural network inference and supports hardware-friendly deployment. The core process includes preprocessing, depth sorting, and volumetric rendering. Despite lower computation than neural radiance fields, it still faces frequent memory access, redundant operations, and bandwidth bottlenecks. To address these challenges, researchers have recently proposed various hardware acceleration solutions for 3DGS. They design low-power accelerators for the full rendering pipeline through joint algorithm-architecture optimization. This paper reviews the basic principles of the method and analyzes key ideas in recent hardware designs. It compares different strategies in module division, data flow control, and memory scheduling. The paper also discusses current challenges in pipeline flexibility, heterogeneous resource use, and system integration. Future work may explore efficient dataflow, adaptive scheduling, and scalable architectures for real-time rendering on edge platforms.