The amount of produced data required to be visualized and analyzed has grown year by year, and the traditional approach of using larger computational resources or exploiting task and data parallelism seems to have reached its limit. Therefore, a new paradigm for large-scale data visualization becomes highly desired, and in this paper, we propose a new and optimized visualization pipeline which uses a point rendering-based early visibility testing for reducing the unnecessary rendering in the final stage of the visualization pipeline, and we named this technique "Early Visibility Test Point Rendering". In a densely populated polygonal mesh scene, where multiple triangles may cover a single pixel, unnecessary and wasteful rendering will occur in the final stage of the traditional visualization pipeline, that is, during the rasterization process. Therefore, we propose an alternative visualization pipeline by introducing the "Early Visibility Test Point Rendering" for selecting only the visible polygonal elements for a given visualization scene. This visibility testing can be executed on the CPU side, and only the visible polygonal elements are needed to be sent to the GPU for an optimized rendering. We could verify the effectiveness of our proposed approach by using synthetic datasets, and also a real-world large-scale simulation result.