This is my take on Peter Shirley's Ray Tracing in One Weekend book series.
This project is using a compute shader dispatched on the GPU, instead of C++ code executed on the CPU, to render the scene. In order to do this, I implemented a heavily simplified version of the C++ code in GLSL (the shader language). This shader code is then dispatched to the GPU using the Vulkan API, where it runs in parallel to render the image.
This is the second part of my 3 project series. Before this project, I followed Peter Shirley' Ray Tracing series and wrote a multi-threaded CPU ray tracer. After this GPU ray tracing implementation using a compute shader, I made a third version using the dedicated Vulkan Ray Tracing extension instead of a compute shader, to speed up the rendering process even further. The extension allows accessing the dedicated Ray Accelerators in the new AMD RDNA 2 GPUs or the dedicated RT Cores in NVIDIA's RTX graphics cards. The performance differences are compared below.
The performance was measured on the same scene (see image above) with the same amount of objects, the same recursive depth, the same resolution (1920 x 1080). The measured times are averaged over multiple runs.
Reference system: AMD Ryzen 9 5900X (12 Cores / 24 Threads) | AMD Radeon RX 6800 XT
| CPU Ray Tracing | GPU Ray Tracing (Compute Shader) | GPU Ray Tracing (Vulkan RT extension) | |
|---|---|---|---|
| 1 sample / pixel | ~ 3,800 ms | 21.5 ms | 1.25 ms |
| 10,000 samples / pixel | ~ 10.5 h (extrapolated) | 215 s | 12.5 s |