This paper presents a comparative study of the Marching Cubes techniques and describes their optimised implementations for a 3D videoconferencing system with the aim of obtaining maximum efficiency and flexibility. Different approaches in the CPU and in the GPU are compared in order to analyze the influence of the rendering process in the reconstruction algorithm. To do this, several 3D scenes were captured, codified and sent to a real-time rendering system that reconstructs 3D models at high speed.