Towards Real-Time Path Tracing

Cornell Box,1k*1k resolution, 4096spp, 20traceDepth

Environment

• cuda/11.7.0-7sdye3
• Nvidia V100 GPU

Getting Started

Linux or Mac

./scripts/setup.sh

./scripts/run.sh cpu_path_tracer 
# or
./scripts/run.sh cpu_path_tracer <SPP>
# or
./scripts/run.sh cpu_path_tracer <SPP> <MaxDepth>

Windows

./scripts/setup.bat

./scripts/run.bat cpu_path_tracer 
# or
./scripts/run.bat cpu_path_tracer <SPP>
# or
./scripts/run.bat cpu_path_tracer <SPP> <MaxDepth>

Now, view the output image in the out directory at root.

Parameters

# cpu path tracer, spp=4096, traceDepth=20, openMP threads=256
build/cpu_path_tracer/cpu_path_tracer 4096 20 256

# cuda path tracer, spp=4096, traceDepth=20, cuda threads=256, Singlekernel Mode
build/gpu_path_tracer/gpu_path_tracer 4096 20 256 1
# cuda path tracer, spp=4096, traceDepth=20, cuda threads=256, StreamCompaction Mode
build/gpu_path_tracer/gpu_path_tracer 4096 20 256 2

Features

Multiple Importance Sampling

MIS

Roadmap

• CPU Path Tracer
• Anti-Aliasing by Jittered Sampling
• Construct GPU-friendly BVH
• Self-defined material description file

GPU Parallelism

• Naive GPU Path Tracer
  • Thread-per-pixel parallelism (# of threads = # of pixels)
• GPU Path Tracer
  • Thread-per-path parallelism (# of threads = # of paths = # of pixels * SPP)
• Work-Efficient GPU Path Tracer
  • Thread-per-ray parallelism (fixed # of threads)

Rendering

• Naive Diffuse Surfaces
• Lambert Material
• Metal Material
• Dielectric Material

Acceleration Structures

• Naive BVH
• SAH BVH

Future Work

• Building BVH on GPU (LBVH https://developer.nvidia.com/blog/thinking-parallel-part-iii-tree-construction-gpu/)