Performance Evaluation of KD-Tree-Optimized Ray Tracing on CPU and GPU

Abstract

Ray tracing has emerged as a powerful rendering technique for generating photorealistic images in scientific visualization, animation, and interactive graphics. This study evaluates and compares the performance of CPU and GPU implementations of a custom ray tracing algorithm without relying on third-party libraries. The CPU version was developed in MATLAB with sequential execution, while the GPU implementation used CUDA with parallel thread processing. Benchmark tests were conducted across resolutions (11×11 to 1000×1000) and scene complexities (1 to 9 spheres). Results show that CPUs outperform GPUs at low resolutions due to lower memory overhead, while GPUs become significantly faster at higher resolutions and complex scenes. Performance profiling reveals GPU execution benefits from scalable parallelism but exhibits variability at smaller workloads. This paper highlights the trade-offs between CPU and GPU-based ray tracing and offers guidance for selecting appropriate hardware based on rendering requirements. Future work includes CPU parallelization, hybrid processing models, and hardware acceleration through VLSI