Java程序辅导

Advanced Computer 
Graphics
(CS & SE 233.420)
Lecture 7 
CREDITS
• Bill Mark, “NVIDIA Programmable Graphics Technology,” 
SIGGRAPH 2002 Course.
• David Kirk, “GPUs and CPUs:The Uneasy Alliance”, Panel 
Presentation, ACM Workshop on General Purpose Computing on 
Graphics Processors.
• David Kirk, “The Future: Programmable GPUs and Cinematic 
Computing”, NVidia Developer Documents, 
developer.nvidia.com/docs/io/4106/Technology_Directions.pdf -Doug James
• GPU History taken from CG Tutorial Book
• www.nvidia.com
• Teaching CG powerpoint presentation provided by NVIDIA
• ACM Workshop on General Purpose Computing on Graphics 
Processors presentations, 
http://www.cs.unc.edu/Events/Conferences/GP2/program.shtml
Graphics 
Processing 
Units 
History
Processors
• The processor is the heart of any normal computer. 
– Intel Pentium, a 64-bit AMD Opteron, or any of the many 
other brands and types of processors
• Most microprocessors are central processing units 
(CPUs)
– Complete computation engines that is fabricated on a 
single chip 
– General purpose and execute applications written in 
general purpose languages, such as C or Java
• Until a few years ago, graphics programmers options
– CPU to process all the transformation and rasterization 
algorithms 
– Very expensive hardware
Graphics Processing Unit 
(GPU)
• Technically
– a GPU is a single chip processor with integrated 
transform, lighting, triangle setup/clipping, and 
rendering engines that is capable of processing a 
minimum of 10 million polygons per second NVIDIA
• In practice
– computationally intensive transform and lighting 
calculations are now offloaded from the CPU onto the 
GPU
GPUs
• Designed is specialized for graphics tasks
• Process 
– tens of millions of vertices per second 
– rasterize hundreds of millions of fragments per 
second
• Process certain tasks much faster than CPUs
– Dot Products
– Vector-Matrix calculations
Pre-GPU
• Graphics acceleration relied on 
– Integrated graphics architectures 
– Specialized and expensive graphics hardware 
pipeline
• Silicon Graphics
• Evans & Sutherland
• These architectures introduced concepts that 
are still relevant in modern graphics
– Vertex transformations, texture mappings
First Generation 
(pre-1998) 
• NVIDIA TNT2, ATI Rage, 3dfx, Voodoo3
• Capabilities
– Rasterizing pre-transformed triangles
– Applying one or two textures
– Implement the DirectX 6 feature set
– When running 2D and 3D applications 
• Completely relieve the CPU from updating individual pixels
• Limitations
– Lack the ability to transform 3D vertices
• Still transformed on the CPU
– Limited set of math operations for combining textures
Second Generation
(1999 – 2000) 
• NVIDIA GeForce 256, GeForce2, ATI Radeon 7500, 
S3 Savage 3D
• Capabilites
– Offloaded  3d Vertex  Transformation & Lighting from the CPU
– Fast vertex transformation now capable on a PC
– Can implement OpenGL and DirectX 7 feature set in Hardware
– Math ops for combing textures and coloring pixels expanded
• Cube map textures
• Signed math operations
• Limitations
– Architecture configurable, but not truly programmable
– Still have a limited set of math operations for combining textures
Third Generation 
(2001)
• NVIDIA’s GeForce3 and GeForce4 Ti, Microsoft’s 
Xbox, and ATI’s Radeon 8500 
• Capabilities
– Provides vertex programmability rather than merely offering more
configurability
• These GPUs let the application specify a sequence of 
instructions for processing vertices
– DirectX 8 pixel shaders and various vendor-specific OpenGL 
extensions expose this generation’s fragment-level configurability
• Limitations
– Pixel-level configurability is available, but these modes are not 
powerful enough to be considered truly programmable.
– Because these GPUs support vertex programmability but lack 
true pixel programmability, this generation is transitional 
Fourth Generation 
(2002-2004)
• GeForce FX family with the CineFX architecture and 
ATI’s Radeon 9700 
• Capabilities
– Provide both vertex-level and pixel-level programmability. 
• This level of programmability opens up the possibility of 
offloading complex vertex transformation and pixel-shading 
operations from the CPU to the GPU. 
– DirectX 9 and various OpenGL extensions expose the 
vertex-level and pixel-level programmability of these GPUs. 
Fifth Generation 
(2004 - )
• GeForce 6 with PCI-Express bus architecture
• Capabilities
– New features include GDDR3 memory 
– Compatibility with the new Pixel Shader 3.0 programming model 
– Microsoft DirectX 9.0 Shader Model 3.0 features set 
– An on-chip video processing engine
• Allows for high-definition video and DVD playback
– 8 operations per pixel at a 60th of a second 
– Support for OpenEXR
• The open standard for filtering in graphics as well as support for 
MPEG encode and decode, as well as support for Windows 
Media Video 9
• The GPU's loops and branches are also programmable and are 
able to write video code to the chip
Transistor Count of CPU & 
GPU Processors
Year Product Name Transistors
1995 Pentium® Pro Processor 5.5 Million
1997 Pentium® II Processor 7.5 Million
1998 Pentium® II Xeon Processor 7.5 Million
1999 Celeron Processor 7.5 Million
1999 Pentium® III Processor 9.5 Million
1999 Pentium® III Xeon Processor 9.5 Million
1999 Pentium® III E 28.1 Million
2000 Pentium® 4 42 Million
2001 GeForce 3 57 Million
2002 GeForce4 Ti 63 Million
2003 GeForceFX 125 Million
2004 GeForce6 220 Million
PCI Express™ Bus 
Architecture
• Interface eliminates the bottlenecks caused by 
previous bus architectures 
• Allows for maximum system performance in a multi-
GPU configuration
• Double the bandwidth of the AGP 8X graphics bus
• 4GB per second in both upstream and downstream 
data transfers
• Opens the door to a truly parallel graphics bus 
architecture
Programmable 
GPU Pipeline 
Evolution
Graphics Pipeline
Vertex
Transformation
Primitive Assembly 
And Rasterization
Fragment Texturing 
and Coloring
Raster
Operations
Vertices Transformed
Vertices
Pixel Positions
Vertex
Connectivity Colored 
Fragments
Pixel
UpdatesFragments
GPU Programmable 
Pipeline
Application VertexProcessor
Fragment
Processor
Assembly &
Rasterization
Framebuffer
Operations
Framebuffer
GPUCPU
Textures
Programmable 
Components
Fifth Generation
NVidia GeForce 6 Pipeline
vertex
setup
rasterizer
pixel
texture
image
programmable per-pixel 
texture, and fp16 blending
programmable vertex
processing (fp32)
programmable per-
pixel math (fp32)
polygon
polygon setup,
culling, rasterization
Z-buf, fp16 blending,
anti-alias (MRT) memory
So what does this 
mean in terms to 
true graphics 
capability?
Nvidia Dawn Demo
GeForce FX 
Demonstration Video
• Two key vertex shaders drive her 
motion:
– Branching skeletal shader 
• the body mesh is driven by several different 
combinations of internal bones
– Blend shape shader 
• deforms her face based on control 
parameters. 
• Skin Shader 
– A complex combination of color maps, 
specular maps, and blood characteristic maps 
to produce very realistic skin. 
– Lighting subtleties are accomplished with a 
series of cube maps for diffuse specular and 
"highlight" skin lighting. 
• Wing Shader  
– A translucent shader is used for the wings. 
– Modifies both the reflected color off the wings as 
well as the amount of light passing through the 
wings based on viewing and light angles.
Video courtesy of Nvidia and CG Tutorial CD
http://www.nzone.com/object/nzone_downloads_nvidia.html
Dawn.mpg
Nvidia Toys
GeForce FX 5800 
Demonstration Video
• Cinematic Camera Effects
– Pixel shaders are used to simulate 
camera effects like depth of field and 
full-scene blurring from an auto-focus 
lens
• Realistic Material Shaders
– Special pixel shaders add realism to 
the toy models in the scene
– Special plastic shader for the tank, 
robot and other plastic models. 
– Painted wood shader is used for the 
wood blocks
– Brushed metal shader for the flying 
saucer
Video courtesy of Nvidia and CG Tutorial CD
http://www.nzone.com/object/nzone_downloads_nvidia.html
Nvidia Time Machine
GeForce FX 5800 
Demonstration Video
• Time-based Shaders 
– Each of the aging materials 
has a single pixel shader 
associated with it. 
– Shaders use a variety of 
texture map inputs (color, 
bump, specular, reflection, 
surface reflectivity, and reveal 
maps) to produce a seamless 
transition of surface material 
effects over time.
Video courtesy of Nvidia and CG Tutorial CD
http://www.nzone.com/object/nzone_downloads_nvidia.html
NVIDIA Nalu
GeForce6 Mascot
• Dense hair is simulated in real time and lit by a 
technique called "deep shadows" where the topmost 
hairs glow brightly from exposure to the light, while the 
lower hairs are darker.
• Her skin is lit by the light refracted through the water's 
surface, her body and hair casting soft shadows on her 
as she swims.
• Soft shafts of light filter down from the surface and are 
blocked by her silhouette using world-class, render-to-
texture capabilities.
• Her high-resolution skin transitions into a highly detailed 
scale shader that features the same soft shadowing as 
the skin, but adds a more noticeable bumpmap, 
iridescence, and bio-luminescence.
• The final render pass (19 in all) provides a soft glow that 
allows the bright light on her hair and skin to bloom on 
the screen.
• See GeForce 6 Series Demo Collage from web page 
below
Video courtesy of Nvidia and CG Tutorial CD
http://www.nzone.com/object/nzone_downloads_nvidia.html
How do you
Program a GPU?
Need the Right Hardware 
• Anything greater that a GeForce 4
– Know what you can do with the card
– Make sure you have the right driver
• Bus architecture will affect how you data from 
CPU to GPU 
– AGP 4x is more limiting that 8x
– PCI-Express quickly making AGP card/mother 
boards obsolete
• NVIDIA leaders in the Programmable GPU 
card development
Need a Programming 
Language or API
• Programming powerful hardware with assembly 
code is hard
• GeForce 6 supports programs more than 1,000 
assembly instructions long
• Programmers need the benefits of a high-level 
language:
– Easier programming
– Easier code reuse
– Easier debugging
• Available APIs or Programming Languages
– OpenGL, OpenGL Shader Language, CG
Low-Level APIs
• Low-level APIs
– Similar to assembler language
• Close to hardware functionality
• Input: Vertex/fragment attributes
• Output: new vertex/fragment attributes
• Sequence of instructions on registers
• Platform dependent
• Current Low-level APIs
– OpenGL extensions (OpenGL 2.0)
• GL_ARB_vertex_program
• GL_ARB_fragment_program
• DirectX: 
– Vertex Shader, Pixel Shader
CG - The High-Level 
Language for Graphics
• Cg is an open-source high-level shading language to 
make graphics programming faster and easier
• Cg replaces assembly code with a C-like language 
and a compiler
• Cg is cross-API (OpenGL & DirectX) and cross-
platform (Windows, Linux, and Mac OS)
• Cg is a key enabler of cinematic computing
• http://developer.nvidia.com/CgTutorial/
OpenGL 2.0 Shading 
Language
• A high-level procedural shading language for 
OpenGL
– Part of the core OpenGL 2.0 specification
• Designed to allow application programmers to create 
shaders for programmable vertex processing and 
fragment processing
– Allows developers to take total control over the most 
important stages of the graphics-processing pipeline
• Based on ANSI C
• http://www.opengl.org/documentation/oglsl.html
Books
• OpenGL® Shading Language, 
Randi J. Rost
• Cg Tutorial, The: The Definitive 
Guide to Programmable Real-Time 
Graphics,  Randima Fernando, 
Mark J. Kilgard.
• GPU Gems: Programming 
Techniques, Tips, and Tricks for 
Real-Time Graphics, Randima
Fernando
• NVIDIA GPU Programming Guide, 
http://developer.nvidia.com/object/gpu_programming_guide.html
Project, Paper, and 
Presentation
Project
• Focus on the aim of your project
• Main point about these projects is to gain 
knowledge in a specific area of computer 
graphics that is of interest to you
• It is up to you to make the project interesting
• Be practical
Paper
• Conference style paper
– Web page provides link to the ACM/Siggraph paper 
submission
– 6-8 pages
• Suggested format
– Abstract (do this last)
1. Introduction
2. Background
3. The Model
4. Simulation Results (very dry)
a. List graphics card, compilers, code (OpenGL 1.?, C/C++, CPU, glut Libs)
5. Discussion (of results)
6. Conclusion
7. Acknowledgments
8. References
Presentation
• 10 minutes (5 Minutes for questions)
– 10 slides max !!!
• Format
– Tell ‘em what your are going to say
• Introduction
– Say it
• Background, Model, Simulation, Discussion
– Tell ‘em what you said
• Conclusion
• Demonstration
– Go over to the lab and demonstrate your code
Due Date
• Week of 13-17 June
– Need 3 hours for presentations
• Tues - 14 Jun, 2pm-5pm
– Need 3 hours for demonstrations
• Thur - 16 Jun, 2pm-5pm
• Papers/Web page due at beginning of 
demonstrations