Posted by Thomas
McGuire on August 2, 2001
Manufacturer: Videologic Product:
The Vivid! XS features an S-Video output jack for TV Output,
although you also get a S-Video to Composite converter cable
with the Vivid! XS to enable the use of TV Output should a
S-Video connection be unavailable.
Tile Based Deferred Rendering
Before moving into further details about the Vivid! XSí
performance it would be wise to talk a bit about Tile Based
Deferred Rendering. Probably the 2 most important factors in
determining a Graphics cards performance would be that of effective
Memory bandwidth & Overdraw.
In recent times we've seen that Graphics Cards have become
increasingly memory bandwidth limited. Last year we had the
Geforce 2 experiencing this & this year we have the
Geforce 2 MX 2/400s in a similar situation. ATI have
developed Hyper-Z in the Radeon to help alleviate this
problem, while the Geforce 3 has Z-Occlusion Culling for
similar reasons. Memory bandwidth is the power needed to
render your Games (Or other 3D accelerated Applications),
the greater your bandwidth, the greater the expected
performance. In a way it's much like your Internet
connection - A person with Cable is going to have a whole
lot more bandwidth at his disposal than someone with a V90
Overdraw. This wastes memory bandwidth & processing
power rendering pixels that arenít visible. As games are
becoming more complex the level of overdraw is increasing
rather drastically, hence the amount of bandwidth required
(& more importantly, wasted) is also increasing
at a similar rate. Some may well argue that developers are
to blame for this problem (The Quake Engines are
semi-renowned for their low amounts of overdraw) but
thatís not really an easy to solve problem.
Tile Based (Deferred) Rendering.
The Kyro2 divides the screen into multiple tiles (Hence why
itís called Tile based rendering) of 32x16 pixels. The
number of tiles created varies with screen resolution, e.g.
1536 tiles are created at 1024*768, while only 600 tiles are
at 640*480. Each of these tiles is then tested by the
On-chip Tile-Buffer to determine whether or not the pixels
are visible. Should they not be visible then they are
disregarded & nothing more need be done with them in the
rendering process. The diagram beneath from the PowerVR TBR
Whitepaper illustrates this rather nicely.
As you can see, the tile-based solution inevitably only
renders foreground objects while traditional 3D renderers
are also rendering non-visible background objects as well.
The Deferred part refers to texturing.
Only once pixels have been verified as being visible by the
On-chip Tile-Buffer are textures applied to them. As such
textures arenít loaded unless they are going to be seen.
This saves memory bandwidth as a result, which aids overall