 |
|
|
|
In the Forums... |
Posted: June 22, 2001 Written by: Dustin Jones Tile Based Rendering With the way the video card market has been proceeding, with the focus on raw power and huge clock speeds, it would seem like the lowly 175MHz core and memory speed wouldn't result in GeForce 2 GTS speed, instead perhaps framerates like those of a TNT2 or GeForce 2 MX at best. This is not the case, however, because PowerVR has re-introduced a unique way of rendering the game scene called Tile-Based Rendering (also known as Hidden Surface Removal). Tile-Based Rendering is PowerVR's justification of the 175Mhz core and 175MHz SDR memory on the Kyro II. As you may or may not know, a scene of a game is made up of pixels and polygons (among other things). There can be tens of thousands of polygons in just one deathmatch session in Quake III. The way most video cards take these pixels and polygons and turn them into frames of an animation is by drawing and filling the entire scene, and texturing all the thousands of polygons before rendering the next frame. This is type of rendering is known as 'Immediate Mode Rendering'. When you think that it takes 30-40 frames per second to get a game running fairly smoothly, that's a lot of polygons being rendered in one second, and thus, a whole lot of rendering power needed. Doesn't sound like a very efficient way to render does it? Well, PowerVR took a look at this and thought the same thing, so they decided to resurrect the fairly old technology of Tile-Based Rendering, which was first seen in the PowerVR PCX2 chipset. That's right, Tile-Based Rendering is far from a new technology, it's actually been around for quite some time, but was not used for currently unknown reasons. So how does Tile-Based Rendering work? Well, as the name states, Tile-Based Rendering processes the scene using grouped tiles instead of processing each polygon at a time. The scene information is directed to the graphics processor, which in turn, cuts the scene up into a number of small tiles. Each tile is then analyzed and using the built-in 24-bit Z-Buffer, parts of each tile that aren't going to be visible in the final scene are Z-Buffered out and not rendered. The Tile-Based system does not apply textures to these non-visible triangles. This saves a tremendous amount of memory bandwidth, as an excessive amount of texels (textured pixels) do not need to be drawn, and then later eliminated at the final scene output. This is called 'Overdraw', and is typical of most modern 3D accelerators, thus the need for massive clock speeds and power. Tile-Based Rendering instead stores the tiles in the on-chip 'Tile Buffer', where texturing and blending then takes place. Think of it this way; you're standing in map in a game and looking at giant monster blocking your view of your way out of the map. You can't see the exit, but you know it's there. Aha, but it's really not, because Tile-Based Rendering has not rendered it and thus, there is nothing behind that monster. Since this is done very transparently, there is no indication that parts of the scene are missing at all, and the game will turn out exactly like its Immediate Mode rendered counter-parts. Now that you know how Tile-Based Rendering works, perhaps the lowly clock speed of the Kryo II will begin to look a little better, seeing as how the Tile-Based system essentially eliminates overdraw and makes for a much more efficient use of the memory bandwidth, making the need for huge amounts of raw speed and fill rate non-existent. Essentially, 500MHz would be a waste of power and heat on the Kyro II, as it just simply does not need it. |
||
|
| |||
|---|---|---|---|