During the last couple of years, point sets have emerged as a new standard for the representation of largely detailed models. This is partly due to the fact that range scanning devices are becoming a fast and economical way to capture dense point clouds. Traditional rendering systems are impractical when a single polygonal primitive contributes less than a pixel during rendering. We present a data distribution strategy for parallel pointbased rendering, using a cluster of PCs as target platform. We describe a data-structure and a system architecture, which allows for decoupling the point-data from the computational work. This strategy enables both a balanced workload as well as no full data replication on each node. We exploit frame-to-frame coherence to make our system scalable. The system renders high-resolution images from high complex data sets at interactive frame rates. To our knowledge parallel point-based rendering has not been investigated in the past. Our results indicate the feasibility of sort-first parallelization applied to point-based rendering.