Function

Stream processing is used to accomplish faster computational speeds for large operations by distributing a single, specific task over a large quantity of stream processors. On modern graphics cards, this most often involves the rendering of a high-resolution image or computation of in-game physics.

Methods

There are two major manufacturers of modern graphics cards: Nvidia and ATI (a subsidiary of AMD). Each manufacturer uses a different architecture. Nvidia cards use stream processors sequentially, with each individual processor being specific in its use and assigned individual tasks. ATI groups the stream processors into units of six, with the first four in each six given a task and the others used to process floating point instructions and data checking. This is why ATI cards often use several times the number of stream processors over that of equivalent Nvidia products.

Benefits

The use of stream processing by GPUs allows for each operation to be divided into as many parts as necessary to complete the task efficiently. The results of this can be seen in increased frame rates, reduced screen tearing and improved load times.

Limitations

GPU stream processors can only perform one type of operation at a time. For example, you could use your GPU to encode high-definition video or render 3-D images, but each stream processor cannot perform both operations simultaneously.

Applications

The use of stream processors in modern graphics cards has expanded the GPU's potential beyond accelerating the display of graphics. Weather forecasting, physics simulation, cryptography and molecular modeling are just a few of the things that modern GPUs are used for today.

Source:

General Purpose Computing on Graphics Hardware

Nvidia Developer Zone

AMD Developer Central

Resource:

Merrimac - Stanford University Stream Computing Project

General Purpose Computing on Graphics Hardware