Abstract

Rapid development in the field of computer graphics over the last 40 years has brought forth different techniques to render scenes. Rasterization is today’s most widely used technique, which in its most basic form sequentially draws thousands of polygons and applies texture on them. Ray tracing is an alternative method that mimics light transport by using rays to sample a scene in memory and render the color found at each ray’s scene intersection point. Although mainstream hardware directly supports rasterization, ray tracing would be the preferred technique due to its ability to produce highly crisp and realistic graphics, if hardware were not a limitation. Making an immediate hardware transition from rasterization to ray tracing would have a severe impact on the computer graphics industry since it would require redevelopment of existing 3D graphics-employing software, so any transition to ray tracing would be gradual. Previous efforts to perform ray tracing on mainstream rasterizing hardware platforms with a single processor have performed poorly. This thesis explores how a multiple GPGPU system can be used to render scenes via ray tracing. A ray tracing engine and API groundwork was developed using NVIDIA’s CUDA (Compute Unified Device Architecture) GPGPU programming environment and was used to evaluate performance scalability across a multi-GPGPU system. This engine supports triangle, sphere, disc, rectangle, and torus rendering. It also allows independent activation of graphics features including procedural texturing, Phong illumination, reflections, translucency, and shadows. Correctness of rendered images validates the ray traced results, and timing of rendered scenes benchmarks performance. The main test scene contains all object types, has a total of 32 Abstract objects, and applies all graphics features. Ray tracing this scene using two GPGPUs outperformed the single-GPGPU and single-CPU systems, yielding respective speedups of up to 1.8 and 31.25. The results demonstrate how much potential exists in treating a modern dual-GPU architecture as a dual-GPGPU system in order to facilitate a transition from rasterization to ray tracing.

Library of Congress Subject Headings

Rendering (Computer graphics); Ray tracing algorithms; Computer graphics; Real-time data processing; Computer architecture

Publication Date

2-1-2009

Document Type

Thesis

Department, Program, or Center

Computer Engineering (KGCOE)

Advisor

Shaaban, Muhammad

Advisor/Committee Member

Geigel, Joe

Comments

Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works. Physical copy available through RIT's The Wallace Library at: T385 .U77 2009

Campus

RIT – Main Campus

Download

Share

COinS