Linchuan Chen. 图形处理器( Graphics Processing Unit ),...

Preview:

Citation preview

Graphics Processing Units

Linchuan Chen

Definition and Functionalities 图形处理器(Graphics Processing

Unit),是一种专门用来处理在个人电脑、工作站或游戏机上图像运算工作的微处理器。

图形处理器使显卡减少了对中央处理器的依赖,并分担了部分原本是由中央处理器所担当的工作

Efficient at manipulating computer graphicshighly parallel structure more effective than general purpose CPUs for

processing of large blocks of data in parallel.

Types of GPUs

集成显卡 (Integrated GPUs)

Types of GPUs

独立显卡 (Discrete GPUs)

GPU Companies

Intel (49.4%)Mainly integrated GPUs

Nvidia (27.8%)The world's first 'GPU‘, GeForce 256,

October 11, 1999 ATI (20.6%) acquired by AMD in 2006

Released its first Radeon 9700 GPU in 2002

Nvidia and ATI control 100% for Discrete GPU market.

How Were Computer Graphics Processed before GPUs Emerged In 90s, 2D GUIs were supported by graphic chips Early – Mid 90s, 3D Games Appeared

3D real time processing were CPU-assisted Low 3D Performance Leads to Hardware-

accelerated 3D graphics. e.g.  S3 ViRGE, ATI Rage,and Matrox Mystique. They are 2D GUI graphic chips with 3D features bolted

on Microsoft provided strict one-to-one support of

hardware for Direct3D. OpenGL is cross-platform The world first GPU (GeForce 256) supports hardware

3D processing

Graphic-only GPUs

After 2000GPUs allow users to do shading before

displaying graphics to the screenNvidia released the first shading capable

chip GeForce 3 March 2001ATI released Radeon 9700 in 2002,

which supports shading with looping and floating point math operations

Stream Processing and General Purpose GPUs (GPGPU) Stream Processing

Stream: a set of data, similar to an imageAll input units are processed in a SIMD wayThis concept turns the modern GPU’s shader

pipeline into general-purpose computing power

General Purpose GPUsEspecially benefits massive vector operationsCan yield several orders of magnitude higher

performance than a conventional CPUBecause each GPU incorporates hundreds of

shader cores

Popular GPGPUs

Nvidia GPUsCUDA SDK, OpenCL SDKTesla Series GPUsFermi Cards

AMD (ATI) GPUsUse OpenCLRadeon Series

GPGPU Programming Model

Coupled CPU-GPUs

Integrates CPU and GPU on the Same Chip, and the GPU supports general purpose computation

AMD Fusion (e.g. AMD A3850 APU) Intel Ivy Bridge Series (e.g. I7 3rd

gen) These are in fact a evolution of

integrated graphic cards

Applications of GPGPUs

Dealing with Highly Parallel Computations Ray Tracing, computational fluid dynamics,

weather modeling, data mining. In the top 500 list, three of the top five

fastest supercomputers in the world were CPU-GPU clusters

However, GPUs are not good at handling branches or data retrieving. For tasks like MS word processing, CPUs are still the first choice