This paper proposes RGBD GS-ICP-SLAM, a dense representation SLAM system that leverages 3D Gaussian representation for high-fidelity spatial representation. The method combines Generalized Iterative Closest Point (G-ICP) and 3D Gaussian Splatting (3DGS) to achieve mutual benefits in tracking and mapping. By utilizing a single Gaussian map for both processes, the system minimizes redundant computations and achieves efficient performance. Keyframe selection methods enhance tracking accuracy and mapping quality. Experimental results show that the proposed approach achieves an impressive speed of 107 FPS and superior map quality. The system is evaluated on the Replica and TUM datasets, demonstrating state-of-the-art performance in spatial representation, camera pose estimation, and overall system speed. The method also addresses issues such as scale alignment and local minima in mapping, leading to improved robustness and accuracy. The code is available at the provided GitHub link.This paper proposes RGBD GS-ICP-SLAM, a dense representation SLAM system that leverages 3D Gaussian representation for high-fidelity spatial representation. The method combines Generalized Iterative Closest Point (G-ICP) and 3D Gaussian Splatting (3DGS) to achieve mutual benefits in tracking and mapping. By utilizing a single Gaussian map for both processes, the system minimizes redundant computations and achieves efficient performance. Keyframe selection methods enhance tracking accuracy and mapping quality. Experimental results show that the proposed approach achieves an impressive speed of 107 FPS and superior map quality. The system is evaluated on the Replica and TUM datasets, demonstrating state-of-the-art performance in spatial representation, camera pose estimation, and overall system speed. The method also addresses issues such as scale alignment and local minima in mapping, leading to improved robustness and accuracy. The code is available at the provided GitHub link.