LOSTCIRCUITS

The 3DLabs VPU

For me personally, the term visual processing unit describes an input device of visual stimuli or else the visual pathway from the retina to higher order cortical and non-cortical centers as shown in detail here whereas use of this term to describe an output device somehow mutilates or actually reverses the term's actual meaning to a paradox. For the sake of marketing, we'd have to accept it, though with reservations.

In short, the term VPU was coined to offset the then new P10 from the rest of the pack, namely, nVidia's NV2x and the ATI R200 graphics engine, both of which were cutting edge when they were new but slowly fading away at the time when the P10 family was first introduced. The P10 family was the first processor family to target a consolidation of the flexibility of a CPU with the performance associated and enabled with highly specialized yet rigid graphics processing units. As such, the P10 naturally needs to adhere to the traditional 5 stage pipeline setting required to convert program and user-intervention-generated data into a visible display:

• (1) transform and lighting (T&L) that is creating of the different vertices that make up the corners of the individual triangles and subsequently cull and clip the scene according to the boundaries of the actual display unit as well as calculating the different lighting effects.
• (2) processing an entire scene at once is tedious and uneconomic, therefore, the scenes are broken up into smaller subunits that are easier to digest, this process is called rasterization of the triangles into tiles. The choice of words is somewhat unlucky since it evokes associations with tile-based architecture otherwise known as deferred rendering. To make it short, the P10 does not use tile-based architecture, rather the entire scene is broken up into 8x8 pixel blocks to match the 64 texture shading processors we'll describe further down the road.
• (3) Hidden surface removal: objects that are hidden one behind the other do not need to rendered, and what is foreground and what would be background needs to be determined as early as possible to save work in any subsequent steps.
• (4) After the visible points of the scene are established and the polygons making up the surfaces have been defined in the 2D viewpoint space of the monitor display, the addresses of the individual pixels are calculated and the textures are applied.
• (5) The final step involves dithering, alpha blending (transparency calculations), anti-aliasing, custom formatting and logic operations.
  

  The image processing pipeline details (above, click for larger picture) and the flow of work along the five processing stages (below). illustrations courtesy of 3DLabs.

  

  We will show the different features in greater detail on the next few pages.

  Next Page:    => Step By Step" =>

  If you enjoyed reading this article and found it useful, please consider making a small donation to LostCircuits.
  Thank you!

  ---

  General disclaimer: This page only reflects the author's personal opinion and assumes no responsibility whatsoever regarding any of the contents or any damages that may occur explicitly or implicitly from reading the contents of this site. All names and trademarks mentioned in this review are the exclusive property of the respective parent companies.
  All contents of this site are protected by international copyright laws. Reproduction of the contents even in parts is not allowed except after written permission by the author and referral to this site.
  Copyright 1998 - 2008 LostCircuits