Power VR's tbr vs. zbuffer

[SuperDuper]

What are the advantages and disadvantages of each?

Is depth precision distributed evenly with Power VR's tile based rendering?

 

[argor]

i recomomend that you repost it over at http://forum.beyond3d.com/index.php

 

[SuperDuper]

okay.

 

[daglesj]

Modern cards with tile rendering could do far more than with the brute force render all approach.

When cards require a 800W PSU just for the card itself someone might say "enough is enough!" and decide to go back to tile rendering.

 

[defaultluser]

Modern cards with tile rendering could do far more than with the brute force render all approach.

When cards require a 800W PSU just for the card itself someone might say "enough is enough!" and decide to go back to tile rendering.

Tile Rendering on PVR is nothing special. It's simply an approach to rendering that's easily scalable, and because the view window of each tile is so small, you can leverage a small texture cache and reduce your external memory bandwidth needs.

The hidden surface removal on PCX1/PCX2 was a better approach than a Z-buffer at the time, because unless you had a large number of triangles inside of a tile, it was faster to sort through the list of triangles inside of a tile than do a Z-buffer read at every pixel. In today's high-end cards, Tile Rendering without a Z-buffer would suck, because each tile would contain thousands of triangles, whereas the number of pixels in a tile is fixed at around 1000 or less.

Now that modern cards have their own impressive texture caches and HSR, there's+no benefit to Tile Rendering over any other choice, beyond the scalability option. ATI cards use Tile Rendering for one of their CrossFireX modes, but it uses a traditional Z-buffer, and simply treats each tile as a small view window of the whole scene.

 

[sweetreddyd]

Tile based rendering was awesome for the time, mostly because it allowed power VR cards to perform very agressive hidden surface removal, now hidden surface removal is more or less a part of every modern GPU, so it loses some of it's kick.

It was a very elegant solution to the problem at the time though, and it would be nice to see it make a comeback. If we could get proper load balancing for multi-gpu setups with it, without artifacting and whatnot, it would be pretty cool. Kinda like what lucid is trying to do with the hydra.

 

[Rys]

Note that we still maintain the depth buffer, and doing so is orthogonal to deferring pixel shading.