Parallel Lazy Amplification: Real-Time Procedural Modeling and Rendering of Multi-Terabyte Scenes on a Single PC

Figure 1: Some images of our massive procedural world. Although quite simple, the entire scene has about 8 terabytes. In this paper, we propose a new procedural modeling paradigm called Parallel Lazy Amplification. This paradigm may be under-stood as a combination between two traditional techniques of pro-cedural modeling: data amplification and lazy evaluation. Data amplification consists in pre-synthesizing the whole geometry be-fore viewing. Alternatively, in the lazy evaluation paradigm, the geometry is generated only when it is needed. The central idea of the paradigm that we propose is to pre-synthesize the geometry that will be potentially seen in the near future and keep it on a cache. A visibility prefetching algorithm determines which models should be generated and which will be discarded. The generation of models is done in parallel with the visualization, without interrupting the sys-tem. We implemented a prototype of this paradigm and some initial experiments with a procedural scene of about 8 terabytes showed the feasibility of this new paradigm, especially when performed on multi-core architectures