Spatial Acceleration · Irregular Parallelism · CPU vs OpenMP vs CUDA

Parallel BVH Construction and Traversal for Accelerated Ray Tracing

A project website for our 15-618 work which focuses on Bounding Volume Hierarchies (BVHs) with ray tracing used as the baseline for analyzing and evaluation.

Soham Narendra Joshi · Viren Dodia
Ray / BVH Concept View
Main Page
Rays enter the scene, BVH culls empty space, and traversal narrows the search for intersections.
Overview

What this project is about

In order to analyze build time, traversal time, load balancing and memory behavior the project focuses parallel BVH construction as well as efficient BVH traversal across sequential CPU, OpenMP multi-core CPU and CUDA GPU implemntations using ray tracing sceness of increasing complexity.

Core Focus

BVH first

We are trying to parallelize the construction and traversal of Bounding Volume Hierarchies to improve ray tracing performance. Ray tracing is to evaluate how BVH helps improve the performance.

Parallel Models

CPU and GPU

We examine how various execution models manage workloads by comparing sequential CPU, OpenMP multi-core CPU, and CUDA GPU implementations.

Main Questions

Speed, scaling, behavior

BVH constrution, traversal efficiency, divergence, memory access ptterns, and the tradeoffs between build cost and overall rendering performance.