Available at: https://digitalcommons.calpoly.edu/theses/3294
Date of Award
6-2026
Degree Name
MS in Electrical Engineering
Department/Program
Electrical Engineering
College
College of Engineering
Advisor
Maria Pantoja
Advisor Department
Computer Science
Advisor College
College of Engineering
Abstract
The real-time rendering of large-scale, procedural scenes presents a significant performance challenge for traditional CPU-bound rendering pipelines. The high volume of draw calls and the need for complex culling and level-of-detail management create bottlenecks that limit scene complexity and visual fidelity. This thesis investigates the evolution of GPU-driven rendering paradigms via the Xylem renderer within NVIDIA’s Donut rendering framework as a solution to these challenges.
A comprehensive benchmarking framework is developed to implement and quantitatively analyze three distinct rendering strategies for a procedurally generated, parameterizable, large-scale forest scene. The evaluated pipelines include: (1) traditional instanced rendering, (2) compute-driven indirect rendering pipeline, and (3) meshlet rendering pipeline. Results show the meshlet pipeline surpassing all other pipelines’ frame times for the sample scenes tested against, while the compute pipeline only outperforms the traditional pipeline in certain instances.