CityGaussian (CityGS) is a novel approach for real-time, high-quality large-scale scene rendering using 3D Gaussian Splatting (3DGS). It addresses the challenges of training and rendering large-scale 3DGS by employing a divide-and-conquer training approach and a Level-of-Detail (LoD) strategy. The method first generates a coarse global Gaussian prior through a divide-and-conquer strategy, followed by parallel training of spatially adjacent blocks with adaptive data selection. This ensures efficient training and seamless fusion of different blocks. The LoD strategy dynamically selects necessary Gaussians for fast rendering, reducing computational overhead. Extensive experiments on large-scale scenes demonstrate that CityGS achieves state-of-the-art rendering quality and consistent real-time performance across various scales. The project page is available at https://dekuliutesla.github.io/citygs/.CityGaussian (CityGS) is a novel approach for real-time, high-quality large-scale scene rendering using 3D Gaussian Splatting (3DGS). It addresses the challenges of training and rendering large-scale 3DGS by employing a divide-and-conquer training approach and a Level-of-Detail (LoD) strategy. The method first generates a coarse global Gaussian prior through a divide-and-conquer strategy, followed by parallel training of spatially adjacent blocks with adaptive data selection. This ensures efficient training and seamless fusion of different blocks. The LoD strategy dynamically selects necessary Gaussians for fast rendering, reducing computational overhead. Extensive experiments on large-scale scenes demonstrate that CityGS achieves state-of-the-art rendering quality and consistent real-time performance across various scales. The project page is available at https://dekuliutesla.github.io/citygs/.