RaDe-GS introduces a rasterized method to compute depth and surface normal maps for general Gaussian splats, enhancing 3D shape reconstruction while maintaining training and rendering efficiency. Unlike 2D Gaussian Splatting, which forces Gaussian splats to be planar and results in blurry novel view rendering and noisy 3D shapes, RaDe-GS computes depth maps through rasterization, achieving high-quality 3D reconstruction with comparable accuracy to NeuraLangelo on the DTU dataset. The method computes depth and normal maps by deriving a closed-form solution for light ray intersections with Gaussian splats, leveraging the approximate affine projection to efficiently compute depth values. The depth map is calculated as the median depth among projected Gaussian splats, considering their translucency, while the surface normal map is derived through rasterized computation. RaDe-GS maintains the computational efficiency of Gaussian Splatting and can be integrated into existing Gaussian Splatting-based methods. The method achieves a Chamfer distance error of 0.69 mm after 5-minute training on the DTU dataset, outperforming recent GS-based methods like GOF and 2D GS. RaDe-GS also demonstrates high-quality novel view synthesis, with results comparable to standard 3D GS methods. The method is efficient, reconstructing scenes in about 8.3 minutes on the DTU dataset and 11.5 minutes on the TNT dataset, significantly faster than implicit NeRF-based methods. RaDe-GS is a significant advancement in Gaussian Splatting, offering a novel approach to depth and normal map computation that improves 3D reconstruction accuracy and maintains rendering efficiency.RaDe-GS introduces a rasterized method to compute depth and surface normal maps for general Gaussian splats, enhancing 3D shape reconstruction while maintaining training and rendering efficiency. Unlike 2D Gaussian Splatting, which forces Gaussian splats to be planar and results in blurry novel view rendering and noisy 3D shapes, RaDe-GS computes depth maps through rasterization, achieving high-quality 3D reconstruction with comparable accuracy to NeuraLangelo on the DTU dataset. The method computes depth and normal maps by deriving a closed-form solution for light ray intersections with Gaussian splats, leveraging the approximate affine projection to efficiently compute depth values. The depth map is calculated as the median depth among projected Gaussian splats, considering their translucency, while the surface normal map is derived through rasterized computation. RaDe-GS maintains the computational efficiency of Gaussian Splatting and can be integrated into existing Gaussian Splatting-based methods. The method achieves a Chamfer distance error of 0.69 mm after 5-minute training on the DTU dataset, outperforming recent GS-based methods like GOF and 2D GS. RaDe-GS also demonstrates high-quality novel view synthesis, with results comparable to standard 3D GS methods. The method is efficient, reconstructing scenes in about 8.3 minutes on the DTU dataset and 11.5 minutes on the TNT dataset, significantly faster than implicit NeRF-based methods. RaDe-GS is a significant advancement in Gaussian Splatting, offering a novel approach to depth and normal map computation that improves 3D reconstruction accuracy and maintains rendering efficiency.