GPU.


51 views
Uploaded on:
Description
Equipment gets less expensive, littler, and all the more intense. Parallel architectures create ... from the conventional altered capacity 3D representation pipeline toward an adaptable ...
Transcripts
Slide 1

GPU Graphics Processing Unit

Slide 2

Graphics Pipeline Scene Transformations Lighting & Shading Viewing Transformations GPUs developed as equipment and programming calculations advance Rasterization

Slide 3

Early Graphics Originally, no particular design equipment All handling in programming on CPU, Results transmitted to casing cushion to start with, outer edge cradles later, inward edge supports. CPU Frame cradle Display

Slide 4

Geometry information Transform & lighting Culling, viewpoint isolate, viewport mapping Rasterization Simple texturing Depth test Frame support mixing More nitty gritty pipeline Simple usefulness exchanged to specific equipment.

Slide 5

Geometry information Transform & lighting Culling, point of view gap, viewport mapping Rasterization Simple texturing Depth test Frame cradle mixing Add more usefulness to GPU. Straightforward usefulness exchanged to particular equipment

Slide 6

Fixed capacity GPU pipeline Scene Transformations Pipeline actualized in equipment Each stage does settled errand Tasks are parameterized Inflexible – altered, parameterized capacities Vector-network operations (some parallelism). Lighting & Shading Viewing Transformations GPU CPU Display Rasterization Frame cradle

Slide 7

Technology propels Hardware gets less expensive, littler, and all the more effective Parallel designs create Graphics handling get more complex (ecological mapping, removal mapping, sub-surface dissipating) Need more adaptability in GPUs.

Slide 8

Geometry information Transform & lighting Culling, point of view partition, viewport mapping Rasterization Complex texturing Depth test, alpha test, stencil test Frame cradle mixing Make this programmable: Vertex Shader Make this programmable: Fragment Shader

Slide 9

Vertex shader Graphics frameworks: change over everything to triangles Pass vertices, normals, hues, surface directions to GPU processor GPU: vertex-based calculations, Independent of different vertices Later, gather into triangles.

Slide 10

Fragment shader Fragment is triangle cut to pixel Interpolate values Multiple surfaces, Alpha, stencil, profundity Independent of different parts Blend with substance of casing cradle.

Slide 11

Geometry information Vertex Shader Vertex Shader Vertex Shader Culling, viewpoint isolate, viewport mapping Introduce parallelism: include different units Rasterization Fragment Shader Fragment Shader Fragment Shader Alpha test, profundity test, stencil test Frame cushion mixing

Slide 12

Shading dialect Shade trees - > Pixar\'s Renderman shader

Slide 13

Shader Language Low level (like constructing agent) however abnormal state dialect compilers: nVidia\'s Cg 4 segment drifting point information sort SIMD

Slide 14

Cg: C-based illustrations program Array & structures Flow control Vectors & frameworks No memory assignment, document I/O

Slide 15

Power GPUs have moved far from the customary altered capacity 3D representation pipeline toward an adaptable universally useful computational motor. The crude computational force of a GPU diminutive people that of the most intense CPU, and the crevice is relentlessly extending. GPUs have moved far from the conventional settled capacity 3D representation pipeline toward an adaptable universally useful computational motor

Slide 16

Next: bind together shaders One arrangement of shaders Allocate to either vertices or parts

Slide 17

Pipeline advanced

Slide 18

Evolved pipeline

Slide 19

GPGPU Make GPU more broad – adjust certain sorts of projects to it\'s pipelined, parallel design Single GeForce 8800 chip accomplishes a supported 330 billion skimming point operations for each second (Gflops) on straightforward benchmarks Cost-compelling: illustrations driving interest up, supply up, cost down for GPUs Finding utilizes as a part of non-illustrations applications.

Slide 20

GeForce 8800 GTX

Slide 21

More broad: NVIDIA\'s CUDA More broad information parallel model Decompose crosswise over strings Sharing and correspondence between strings..

Recommended
View more...