GPU design issues
John Danskin
NVIDIA

11/16/05, 1pm
Location TBD

Abstract

Walkthrough of a modern NVIDIA GPU design focussing on current design
issues including: index and vertex fetch, attribute pipelines, clipping,
culling architectures, pixel shader, state changes, texture level of
detail computations, divergent branching, shader caches, fragment
crossbars, architectural memory efficiency, frame-buffer memory
controllers.

Bio

John M. Danskin started working in Computer Graphics as an undergraduate
at UCSC in 1979. After working for a number of companies including Daisy
Systems and Digital Equipment, where John participated in the design and
microcoding of several graphics processors, John went back to school in
1989. John was granted a Ph.D. in Computer Science from Princeton in
1994. John taught at Dartmouth for a few years then rejoined industry in
1997 to work on GPU architecture for Dynamic Pictures and then 3dfx
before joining NVIDIA in 2000. John has contributed to all of the major
NVIDIA GPU families since NV20. John is now a director of architecture
for NVIDIA, responsible for the architecture of an upcoming family of
GPUs.