Design of a parallel accelerator for volume rendering

Abstract

We present the design of a flexible massively parallel accelerator architecture with simple processing elements (PEs) for volume rendering. The underlying parallel computer model is a combination of the SIMD mesh with the instruction systolic array (ISA), an architectural concept suited for easy implementation in very high integration technology. This allows the parallel accelerator unit to be built as a programmable low cost co-processor, that suffices to render volumes with up to 16 million voxels (2563) at 30 frames per second (fps).