This shows you the differences between two versions of the page.

Link to this comparison view

seminars:seminar_11_09_14 [2017/09/20 22:02] (current)
Line 1: Line 1:
 +======Efficient,​ Heterogeneous,​ Parallel Processing: The Design of a Micropolygon Rendering Pipeline======
 +Wednesday Sep. 14, 2011\\
 +Hamerschlag Hall D-210\\
 +4:00 pm\\
 +**[[http://​www.cs.cmu.edu/​~kayvonf/​|Kayvon Fatahalian (CS, CMU)]]**\\
 +CS, CMU\\
 +The modern real-time graphics pipeline is a versatile parallel
 +architecture that accommodates a wide range of graphics techniques.
 +The architecture is implemented by heavily optimized GPUs that employ
 +a mixture of application-programmable and fixed-function processing
 +resources, yet its design lends itself to a simple programming model
 +easily understood by non-expert programmers.
 +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
 +pixel-sized "​micropolygons",​ fundamental system operations such as
 +geometry processing, surface visibility, and shading execute
 +inefficiently under this workload.
 +In this talk I will describe a proposed evolution of the graphics
 +pipeline that increases system efficiency under micropolygon
 +workloads. The redesign includes simultaneous algorithmic,​ GPU
 +hardware, and pipeline abstraction changes that integrate high-quality
 +adaptive tessellation and motion blur, increase rasterization
 +parallelism,​ and reduce shading costs by more than a factor of eight.
 +When making these changes, irregular communication and control-flow
 +are isolated to non-programmable parts of the system, preserving the
 +graphics pipeline'​s simple, implicitly parallel programming model and
 +the throughput-optimized design of the GPU's programmable cores. ​ This
 +"​graphics-style"​ approach to system design is interesting to consider
 +in other compute-intensive domains and in the development of more
 +general parallel computing systems.\\
 +Kayvon Fatahalian is an Assistant Professor in the Computer Science
 +Department at Carnegie Mellon University. His research focuses on the
 +design of efficient parallel systems for computer-intensive applications
 +such as interactive computer graphics.\\
 +**[[seminars| Back to the seminar page]]**