Interactive Distributed Ray Tracing of Highly Complex Models

Memory sharing for interactive ray tracing on clusters

“…Several works by Wald et al [7][8][9] demonstrate interactive ray tracing of large models in both single PC and distributed cluster environments. Their first system [7] traced four rays at a time using SIMD instructions to accelerate the rendering process.…”

“…An additional benefit of this technique is that the data coherence of primary rays is automatically exploited. Other early work by Wald et al [8] addressed the challenges of interactively rendering large, complex models by combining centralized data access and client-side caching of geometry voxels. Their system takes pains to exploit spatial coherence within BSP tree nodes and temporal coherence between subsequent frames.…”

“…This pseudo-static assignment scheme increases hit rates because the data used to render frame t − 1 will likely be needed when rendering frame t. The goal of this approach is similar to the scheduling heuristic described by Wald et al [8].…”

Rendering Complex Scenes on Clusters with Limited Precomputation

“…The RTRT team [Wald and al, 2001a] proposes the first animation of large models by distributing an optimized ray tracing engine on commodity computers [Wald and al, 2001b]. The model is precomputed by the construction of a binary space partition (BSP) tree.…”

“…Like in another study [Wald and al, 2001b], a fixed size of the tiles (32x32 pixels) is set. Results show that parallel efficiency decreases about by 3% and thus validate the contribution of the temporal subdivision method of tiles.…”