How does a GPU work [40 mn of reading - published 5/18/2006 11:25:30 PM - Target : Débutant]

Author

Thibault MICHEL Student-Engineer Supinfo Paris SUPINFO graduate year  2010    Write to the person    All projects of the same author    Mini-CV of the author

3 - The vertex pipeline

3.1 - Definition

According to Wikipédia, a pipeline can be defined as being “a technique of design of processors where the execution of several instructions overlap inside the processor itself”. Obviously, the vertex pipeline will use information of a vertex, and process it for the subsequent procedure of modeling.

Among the two types of pipeline whose complicated working uses matrices, one can distinguish:

- The fixed pipelines have existed for some time in GPU; their working can be summarised in the exploitation of the mathematical data of a vertex by mathematical tools. The coordinates of the vertex are used to make transformations such as translation, rotation.

- The programmable pipelines are even more complex, even though their working is quite similar to that of fixed pipelines. Concretely, their working is quite the same as that of fixed pipelines, but with the addition of an important factor: the programmer can replace the mechanisms specific to the processing of vertices, by his own program, hence the name of “programmable” pipelines.

3.2 - additional Performances and operations

Let us see the performances of the vertex pipeline.
Among all the involved parameters, we can distinguish the bus report, the size and speed of which being given by the manufacturers, contrary to many others. The bus report ensures the transport of data between the memory and the GPU, therefore the larger its size and the higher its speed, then the larger the performance.

Following the vertex pipeline, small operations will be carried out on data, to progress further towards depiction. For your information these operations include clipping, culling and tramage. They all aim at performance by the removal of superfluous data in specific situations, but also at as sharp as possible surfaces (smoothing for instance).

Finally, we obtain at this stage of treatment pixels and information required for the desired depiction. Next important stage: the pixel pipeline.