https://github.com/kevinpruvost/cuda_raytracing

Personal Project on implementing Ray Tracing with CUDA, visualizable on an OpenGL Renderer
https://github.com/kevinpruvost/cuda_raytracing

Last synced: 4 days ago
JSON representation

Personal Project on implementing Ray Tracing with CUDA, visualizable on an OpenGL Renderer

• Host: GitHub
• URL: https://github.com/kevinpruvost/cuda_raytracing
• Owner: kevinpruvost
• Created: 2024-05-17T13:26:49.000Z (8 months ago)
• Default Branch: main
• Last Pushed: 2024-05-23T03:45:03.000Z (8 months ago)
• Last Synced: 2024-05-23T03:50:16.087Z (8 months ago)
• Language: C++
• Size: 13.5 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# CUDA_RayTracing

Personal Project on implementing Ray Tracing with CUDA, visualizable on an OpenGL Renderer

## Resources

[CUDA/OpenGL Interoperability Documentation](https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__OPENGL.html)

[CUDA Architecture Code Matching for different setups](https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/)

[Post on how occupancy influences kernel performances](https://stackoverflow.com/questions/6688534/cuda-dependence-of-kernel-performance-on-occupancy)

[BVH: Bounding Volume Hierarchy](https://en.wikipedia.org/wiki/Bounding_volume_hierarchy)

[Acceleration Techniques (BVH/Kd-Tree/Octree)](https://www.csie.ntu.edu.tw/~cyy/courses/rendering/15fall/lectures/handouts/chap04_acceleration_4up.pdf)

## Notes

- When working with multiple files that need external symbol links, don't forget (on Visual Studio) to activate Properties/Cuda C/C++/Common/Generate Relocatable Device Code.

- I am working with an RTX 3070 Laptop, so the GPU is using Ampere architecture, sm_86 should be used then.

- For the number of threads and blocks, I will just go with 16x16 threads, and a number of blocks based on the number of pixels to render divided by the number of threads, so `(width + block_size - 1) / block_size` and `(height + block_size - 1) / block_size`.

- Use Nvidia NSight Compute either on standalone app or Visual Studio plugin for Performance Profiling on CUDA.

- For projects like this, where a lot of data might be created, we have to manage stack size limit (`        cudaDeviceSetLimit(cudaLimitStackSize, N)`).

- Be careful with the random seed generation also, and give different seeds to each device.

- Stack size on device has a huge impact on performance, be careful.