One central problem in geospatial applications using 3D models is the tradeoff between detail and acquisition cost during acquisition, as well as processing speed during use. Commonly used laser-scanning technology can be used to record spatial data in various levels of detail. Much detail, even on a small scale, requires the complete scan to be conducted at high resolution and leads to long acquisition time, as well as a great amount of data and complex processing. Therefore, we propose a new scheme for the generation of geospatial 3D models that is driven by relevance rather than data. As part of that scheme we present a novel acquisition and analysis workflow, as well as supporting data-models. The workflow includes on-site data evaluation (e.g. quality of the scan) and presentation (e.g. visualization of the quality), which demands fast data processing. Thus, we employ high performance graphics cards (GPGPU) to effectively process and analyze large volumes of LIDAR data. In particular we present a density calculation based on k-nearest-neighbor determination using OpenCL. The presented GPGPU-accelerated workflow enables a fast data acquisition with highly detailed relevant objects and minimal storage requirements.