GPGPU stands for general PUR-float Computation on Graphics processing unit, and meant that the diagram processor for tasks is used, for which it actually was not developed. Under certain conditions hereby an enormous speed increase can be obtained compared with CCUs.

Overview

GPGPU stands for the diagram hardware for "“general PUR-float Computation on Graphics processing unit"” and is one by the newer development in the range, in particular the programmable one pipeline by fragment and Vertex Shader as well as the enormous speed of GPUs necessarily and/or meaningfully appear-ends technology, which implements algorithms for usual problems, which are implemented otherwise on the CCU, on the GPU. The restriction of the GPU on special problems without large administration expense makes it possible to sketch these in such a way that the majority of the transistors for arithmetic operations is used and not for control problems and Caching, as it is on CCUs the case. Thereby an optimal achievement results in the case of GPGPU applications, the one high arithmetic density exhibits (thus algorithms, with which the quotient from implemented arithmetic operations per implemented vintage/writing operation is large).

History

The beginnings of the GPGPU are to equate with the beginnings of the programmable diagram pipeline, thus in the year 2000. Even if today many GPGPU applications e.g. uses individual elements of fixed OD Function pipeline like Z Culling, then the flexibility is not large enough, in order to become fair general tasks. Admittingness surely reached GPGPU by an appropriate course on the SIGGRAPH 2004 ( As a first reference work can probably be regarded GPU GEMS2, which appeared 2005.

Advantages

The advantage of the use of the GPU opposite the CCU is mainly in their larger speed. As comparison: nVIDIA the GeForce 6800 reaches peak of values of 60 GFlopsund 18 GB/s Sequenzielle memory bus range, almost simultaneous the Pentium 4 appeared reaches values of straight once 12 GFlops and 5 GB/s range. Those is partly reached by the high degree at parallelism: the GPU implements SIMD (fragment Shader), MISD (Rasterizer) and MIMD (Vertex Shader), whereby the pipeline works as whole task parallel, since fragment and Vertex Shader can be at the same time implemented. A further advantage is the small price compared with similarly fast other solutions as well as the fact that diagram maps are to be found today in almost any PC.

Disadvantages

Many of the GPU solved task not uniformly and the differences between the manufacturers are specified are larger than with (usual) CCUs. In addition are by certain restrictions (is possible e.g. no Scattering) as well as the parallelism special concepts necessarily. This leads to the fact that for example the minimum complexity for sorting O (n*log (n)) amounts to.

Programming

Since in the current computer organization no mechanism is intended, which makes programs directly on the GPU executable, it is necessary to implement an application of frameworks implemented on the CCU which transacts the appropriate function calls of the diagram map. For example if an application is to be accomplished by fragment Shadern, then must be produced first for this fragment, which is usually reached by the Rendern of a rectangle. Too rendern as well as the diagram map receives to whom those the data, the instruction, a rectangle with activated fragment Shader to the Shader code thereby from the CCU based application of frameworks. The result must likewise become processed by the application of frameworks (e.g. stored), if it is not only to be indicated in form of the computed Framebuffers.

Literature

Pharr, matte: GPU Gems 2: Part IV - General PUR-float Computation on GPUs: A primer. Addison Wesley Publishing company, 2005

Related links

• GPU Gems 2
• gpgpu.org
• IEEE Vizualization 2005
• SIGGRAPH 2005 GPGPU Course
• To GPGPU in the context of the pro min acre Multimedia Hardwareerweiterungen of the AG reactive systems to DO Kaiserslautern

Articles in category "GPGPU"

We found here 3 articles.

Related Websites

We found here 4 related websites.

• GPGPU
  GPGPU stands for General-Purpose computation on GPUs. ... The Accelerator GPGPU programming system from Microsoft Research is now available for download. ...


• GPGPU - Wikipedia, the free encyclopedia
  General-Purpose Computing on Graphics Processing Units (GPGPU, also referred to as GPGP and ... Some improvements were needed before GPGPU became feasible. ...


• GPGPU : Developer Resources
  Here you will find information about GPGPU programming, including links, ... This introductory tutorial teaches five essential GPGPU concepts through the ...


• GPGPU : SIGGRAPH 2004 Course
  GPGPU stands for General-Purpose computation on GPUs. With the increasing programmability of commodity graphics processing units (GPUs), these chips are ...