The visualization of large terrains is a challenging Computer Graphics issue. The number of polygons required to faithfully represent a terrain`s geometry can be too high for real-time visualization. To solve this problem, a multiresolution algorithm is used to feed the graphics processor only with the most important polygons, without visual quality loss. The amount of data is another important problem, as it can easily exceed a computer`s RAM. Thus, a system to manage out-of-core data is also required. The present work proposes a simple and scalable solution to visualize the geometry of large terrains based on three key points: a data structure to represent the terrain in multi-resolution, an efficient visualization system and a data paging and prediction system. Similarly to other works, the system uses a quadtree data structure due to its simplicity, along with the efficiency and the low memory use of an array-based implementation. Each node of the quadtree represents a tile of the terrain. The implementation is divided in two threads, one to manage the tiles and the other for visualization. The tile-management thread is responsible for loading/unloading tiles into/from the memory. This thread uses a camera-movement prediction mechanism to load tiles that can be used in the near future and to remove tiles that probably will not be necessary. The visualization thread is responsible for viewing the terrain, computing the projected error, eliminating tiles that are not visible and balancing the quadtree structure in order to eliminate cracks or T-vertices on the terrain`s surface. The visualization can be made by means of a fidelity-based or a budget-based approach.