Flash Graphics
I designed and
built Lightning, the Flash
Distributed Framebuffer, with John
Owens.
Homan Igehy, Kekoa Proudfoot and I wrote a paper on tolerating memory system latency in a texture caching architecture. Homan and I then extended this work to texture caching in parallel graphics architectures with various rasterization schemes.
Homan Igehy and I developed Pomegranate, a novel parallel hardware architecture for polygon rendering. Pomegranate achieves near-linear scalability at up to 64-way parallelism. Our results are described in a SIGGRAPH 2000 paper.
Gordon Stoll and I took another pass on the original Lightning project to develop Lightning-2 in conjunction with Intel. Our system is described in a SIGGRAPH 2001 paper.
I spent some time working with Greg Humphreys and Ian Buck on scalable cluster rendering. The extension of this system to support the Parallel API is described in a SIGGRAPH 2001 paper.
PhD: Designing Graphics
Architectures Around Scalability and Communication
My dissertation looks at scalability and communication in parallel
graphics architectures. The first half presents a framework and
extended taxonomy of graphics architectures, and the second half
presents Pomegranate, a new
parallel graphics architecture designed for high scalability.
Publications- Lightning-2: A High-Performance Display Subsystem for PC Clusters
- Gordon Stoll, Matthew Eldridge, Dan Patterson, Art Webb,
Steven Berman, Richard Levy, Chris Caywood, Milton Taveira,
Stephen Hunt, and Pat Hanrahan
- to appear in Computer Graphics (SIGGRAPH 2001 Proceedings)
- WireGL: A Scalable Graphics System for Clusters
- Greg Humphreys, Matthew Eldridge, Ian Buck, Gordon Stoll,
Matthew Everett and Pat Hanrahan
- to appear in Computer Graphics (SIGGRAPH 2001 Proceedings)
- Distributed Rendering for Scalable Displays
- Greg Humphreys, Ian Buck, Matthew Eldridge, and Pat Hanrahan
- IEEE Supercomputing 2000
- Parallel Texture Caching
- Homan Igehy, Matthew Eldridge, and Pat Hanrahan
- Proc. 1999 Eurographics / SIGGRAPH Workshop on Graphics Hardware
- Pomegranate: A Fully Scalable Graphics Architecture
- Matthew Eldridge, Homan Igehy, and Pat Hanrahan
- Computer Graphics (SIGGRAPH 2000 Proceedings)
- Prefetching in a Texture Cache Architecture
- Homan Igehy, Matthew Eldridge, and Kekoa Proudfoot
- Proc. 1998 Eurographics / SIGGRAPH Workshop on Graphics Hardware
Master's Thesis: SIMD Column-Parallel Polygon Rendering
I did my Master's Thesis on massively parallel polygon rendering on
SIMD ring architectures, specifically the 1024 processor Princeton
Engine at the David Sarnoff Research
Center. 1-D ring architectures such as Princeton Engine are
easily scaled, but the average interprocessor communication distance
scales linearly with the number of processors, which has serious
impacts on parallel algorithms requiring all-to-all communication,
such as polygon rendering.