Jay E. Steele

Introduction to GPGPU programming

This tutorial will present audience members with the foundations necessary to implement general purpose computations on graphics processing units (GPUs). Initially designed for computer graphics, current GPUs have evolved into programmable, highly parallel, floating point processing units. GPUs are now viewed as inexpensive coprocessors that are ideally suited for many applications beyond computer graphics. In this tutorial, the presenters will provide attendees with the knowledge required to realize the full potential of GPUs by introducing industry standard languages, tools, and techniques. Specifically, this tutorial aims to provide those outside of computer graphics with the basic knowledge required to realize the performance gains offered by todays GPUs on general purpose computations.

A lighting model for fast rendering of forest ecosystems

Real-time rendering of large-scale, forest ecosystems remains a challenging problem, in that important global illumination effects, such as leaf transparency and inter-object light scattering, are difficult to capture, given tight timing constraints and models that typically contain hundreds of millions of primitives. This paper proposes a new lighting model, adapted from a model previously used to light convective clouds and other participating media, together with a distribution of ray processing across multiple GPUs, in order to achieve these global illumination effects while maintaining near real-time performance. The lighting model is based on a lattice-Boltzmann method in which reflectance, transmittance, and absorptance parameters are taken from measurements of real plants. The lighting model is solved as a pre-processing step and requires only seconds on a single GPU. The ray tracing engine uses the well-known short-stack algorithm, due to Horn, Sugerman, Houston, and Hanrahan. Both the pre-processing step and the ray tracing engine make extensive use of NVIDIApsilas compute unified device architecture (CUDA).

Convective clouds

A new technique for rendering convective clouds is suggested. The technique uses two lattice-Boltzmann (LB) models, one for generating the spatial and temporal distribution of water density and the other for photon transport, that is, lighting the water density with correct anisotropic scattering. The common LB structure is easily mapped to parallel execution environments such as a GPU or multiple CPUs connected via the Message Passing Interface (MPI), thereby providing sub-minute execution times on commodity hardware.

Relighting Forest Ecosystems

Real-time cinematic relighting of large, forest ecosystems remains a challenging problem, in that important global illumination effects, such as leaf transparency and inter-object light scattering, are difficult to capture, given tight timing constraints and scenes that typically contain hundreds of millions of primitives. A solution that is based on a lattice-Boltzmann method is suggested. Reflectance, transmittance, and absorptance parameters are taken from measurements of real plants and integrated into a parameterized, dynamic global illumination model. When the model is combined with fast shadow rays, traced on a GPU, near real-time cinematic relighting is achievable for forest scenes containing hundreds of millions of polygons.

Parallel processing flow models on desktop hardware

Numerical solution of any large, three-dimensional fluid flow problem is a computationally intensive task that typically requires supercomputer solution to achieve reasonable execution time. This paper describes an alternative approach, a technique for mapping three-dimensional fluid flow models to low-cost, desktop hardware. The approach is shown to deliver exceptional performance.

Second-order illumination in real-time (student paper)

This paper presents a simple and efficient algorithm for achieving real-time performance on current consumer graphics hardware when rendering complex, dynamic scenes with direct and secondorder diffuse (indirect) illumination. An image space, low-discrepancy sampling technique for positioning point lights is presented. These point lights simulate second-order diffuse illumination throughout the scene. A novel use of negative point lights allows fast approximation of occlusion of second-order diffuse illumination in image space. Finally, an optimization technique is provided that improves frame rates while maintaining image quality by approximating the illumination of all point lights at lower resolutions for less detailed areas of the scene.