Efficient, Heterogeneous Parallel Processing: The Design of a Micropolygon Rendering Pipeline
Speaker: Kayvon Fatahalian , Stanford University.
Date: March 1 2010
Time: 4:00PM to 5:00PM
Location: 32-G449
Host: Arvind and Martin Rinard, MIT
Contact: Francis Doughty, 253-4602, doughty@mit.edu
Relevant URL: Designing systems that are high-performance, power-efficient, and
easily programmable by non-experts is an important problem at all
levels of computing. While recent innovations in parallel hardware,
compilers, and programming abstractions address this challenge in a
general context, the modern real-time graphics pipeline is a versatile
and unique parallel architecture that has achieved similar goals
through the co-design of algorithms, programming interfaces, and
hardware.
A major goal of future graphics systems is rendering geometrically
complex, film-quality scenes at interactive rates. Unfortunately,
current GPU implementations not only require additional compute
capability to handle high-resolution surfaces represented by sub-pixel
"micropolygons", fundamental system operations such as geometry
processing, surface visibility, and shading execute inefficiently
under this workload.
In this talk I will describe an evolution of the graphics pipeline
that significantly increases system efficiency under micropolygon
workloads. The redesign includes algorithmic, GPU hardware, and
pipeline abstraction changes that increase parallelism and eliminate
redundant work. Complex communication and control-flow are isolated to
non-programmable parts of the system, preserving the graphics
pipeline's simple, implicitly parallel programming model as well as
the throughput-optimized design of programmable hardware components.
This "graphics-style" approach to efficient parallel computing is
interesting to consider in other domains and in more general system
design.
Bio:
Kayvon Fatahalian is a Ph.D. candidate in Computer Science at Stanford
University. His research focuses on the design of programming
abstractions and efficient parallel systems for compute-intensive
applications such as interactive graphics. Prior to Stanford, he
earned a B.S. in Computer Science from Carnegie Mellon University.
See other events that are part of CS Special Seminar Series Spring 2010
See other events happening in March 2010
