Path Tracing Optimization in Indiana Jones™

1. Shader Execution Reordering

• Implemented for RayGen shaders to increase the average percentage of active threads per warp.
• SER was added as a shader permutation in the main path-tracing RayGen shader for GPUs supporting SER (GeForce RTX 40 and beyond).
• Coherence hint was used to improve performance.

2. Ray-Tracing Live-State Optimizations

• Identified GLSL variables causing RT live state bytes using Ray Tracing Live State tab in GPU Trace Profiler.
• Spilled ray-tracing live-state bytes were reduced or eliminated to optimize GPU performance.
• Optimization 1: Loop restructuring to reduce RT live state.
• Optimization 2: Using FP16 precision for specific variables.
• RT live-state optimizations reduced RT live state from 222 to 84 bytes, improving GPU time spent on a pass by 15% on an NVIDIA RTX 5080 GPU.

3. Performance Comparison

• Before optimizing RT live state, enabling SER made the pass 11% cheaper on a GeForce RTX 5080 GPU.
• After applying RT live state optimizations, a comparison of SER ON to SER OFF showed improved performance.

4. Additional Optimizations

• Additional RT live-state optimizations were implemented in the GLSL code for further performance enhancements.
• Changes in GLSL code optimized RGB throughput vector of the path tracer.

By implementing Shader Execution Reordering and optimizing Ray-Tracing Live-State, significant improvements in GPU performance were achieved for path tracing in Indiana Jones™. Additional optimizations further enhanced the performance and efficiency of the path tracer in the game.