Tile based rendering breaks up the screen into small tiles and renders each seperatley. For each tile, all the polygons are calculated including depth then only those pixels that are visible get textured. This decreases the required texture memory bandwith to the point that cheaper ram can be used allowing for cheaper cards. Either that or it calculates the polygons pixels, checks depth, then textures visible pixels as the polygons arrive. Either way it does decrease the required texture bandwith just not as much. The Kyron II, succesor to the PowerVR, implements tile based rendering. It does not have hardware transform so the CPU must do the tranforms, this should allow it to send only the polygons visible in the current tile. If the next version does implement hardware transform then I would suspect that it would have to transfer the entire scene for each tile.
Full scene rendering of course renders the scene all at once, the scene data is sent to the card only once per frame. The original method did the texturing before the depth test so there was a lot of wasted texture bandwidth. Newer cards I believe do the depth test before texturing a pixel. If so then tile based rendering offers no significant improvement and in fact could lead to a serious decrease in performance once they start including hardware transform.
Here is a comparison of the Kyron II with a Geforce 3 TechTV | 3D Prophet 4500
posted by TechnoMage at 6:29 PM LINK
