Ray Tracing on the GPU

1
Ray Tracing on the GPU

• Mike Allgyer

2
What is ray tracing?

• Method of generating photo-realistic images
• Simulates light bouncing around an environment
• Traces light from camera to world

3
Ray tracing contd

• Every pixel yields at least one ray (more for
  antialiasing)
• Objects are tested for intersection with ray
• Intersection with shortest distance is drawn
  (closest to camera)

4
Ray tracing contd

• If hit
• If object is reflective or refractive, bend
  light, spawn another ray (recursively)
• Spawn ray from intersection to light source to
  calculate shadows
• Color based on shading, texture, etc.

5
(No Transcript)
6
Why use rasterization?

• Ray tracing is slow
• Scenes can take hours or days to render
• GPUs designed for rasterizing images
• Current CPUs not powerful enough for real-time

7
Ray Tracing vs. Rasterization

• Perfect, easy shadows
• Specular highlights
• Straightforward reflection and refraction
• Not just triangles!

8
CUDA

• New nVidia 8800 GPU
• C compiler available to use GPU as general
  purpose CPU
• My project!

9
Why GPU?

• Floating-point arithmetic
• Massively multi-threaded
• Intel quad-core processor 4 processors
• nVidia 8800 GTX 128 processors
• Ray tracing ideal for parallel processing

10
Parallelization

• 2 ways in ray tracing
• Pixel space
• Object space

11
Parallelizing pixel space

• Each pixel independent
• Can be calculated in any order
• Each thread requires knowledge of entire scene
• GPU 16 kb shared memory very limiting

12
Parallelizing Object Space

• Each thread computes intersection of one object
• Thread only needs knowledge of ray and single
  object.

13
Other Obstacles

• No recursion
• Memory may still be a problem

14
Project Summary

• Real-time ray tracer
• Implement on nVidia 8800 using CUDA
• Parallelize object space

15
Results

• Frame Rate
• Resolution
• Number of objects
• Ideal
• 60 fps
• 1280x1024
• 100,000 triangles
• Well see!

16
Questions?
17
References

• http//en.wikipedia.org/wiki/Ray_tracing
• http//hof.povray.org/office-13.html
• http//www.nvidia.com/page/8800_tech_briefs.html