Parallelization of Ray Tracing Units
Computer Graphics

 
A parallel system of two RTPs (ray tracing processors) driven by CPU. The scene data and structure trees are generated in CPU and made accessible to both RTPs. Each RTP generates a partial image. Both parts are merged by the primary RTP into a full image.

Project Description:
 

The increasing GPGPU and Multicore performance raises expectations for commercial RTP (Ray Tracing Processors) of high photorealism and high interactivity, replacing current rendering technology for videogames. The assumption is that multiple RTP systems, in analogy to multiple GPU (e.g. SLI by nVidia, Hydra by Lucid), will be required to accelerate the overall ray-tracing performance. This project studies a split multiple RTP ray-tracing, by simulating a system of CPU and dual RTP by three separate PCs connected by LAN.



Tools and capabilities:
 

  1. Ray Tracing software (e.g. OpenRT)
  2. 3 PCs interconnected by LAN
  3. Data for several static scenes



Project Goal:
 

In this project we will follow the following steps:

  1. Running the entire scene on primary PC
  2. Running ½ of the scene on primary PC
  3. Splitting the system on three PCs simulating a CPU driving two RTPs (see Fig. 1). The resulting output is displayed on primary PC.
  4. Parameter gathering and analysis
  5. Parameters: memory to RTP bandwidth, inter-RTP communication, required data duplication, overheads, load balance, needed scheduling, possible ways of data split.
Project Details:
 
  1. Supervisor:
    Hovav Gazit
    Phone: 04-8295741
    Email: hovav@ee.technion.ac.il
  2. Field:
    Computer Graphics
  3. Requirements:
    Open minds...
  4. Project status:
    Taken by the students Guy Dor Leferfein and Eli Tsiben
    Visit:
    Project web site
 
Current Projects Archive
CG&M Lab    Contact Us EE Labs EE Department Technion