New single- and multiprocessor models for ray tracing are presented. Important features are (1) the use of custom VLSI building blocks, (2) the use of a modified hierarchical data-structure-based ray tracing algorithm with three disjoint data sets, and (3) scene access through adaptive information broadcasting. A modular design is presented that permits incremental performance enhancement up to two orders of magnitude over conventional minicomputers or workstations. Ray tracing is a surprisingly good application for a shared bus architecture because of the computational complexity of intersecting light rays with graphics objects.