Traditional deferred rendering and "light prepass" are very similar techniques. Especially when you are doing light prepass in a single pass like Cryengine 3 and most other new games do it. I personally consider single pass light prepass as being fully deferred. The only difference really is that LPP saves one tfetch for each overlapping light, but you need to fetch the light accumulation buffer later to finish the light equation. The g-buffers of single pass LPP are identical to "fully" deferred renderers (you just do not use them all in the lighting pass, color/specularity g-buffer is sampled later).
Tiled deferred techniques have been used a lot in PS3 games (the Uncharted 1 was one of the first ones to use it). Majority of PS3 deferred renderers seem to be tiled, as tiling suits PS3 architecture better. Tiling is just a bandwidth optimization. The lighting is calculated exactly the same way as in any other deferred renderer, and the g-buffer contents are identical. Also all the same downsides and upsides of other deferred techniques apply.
Tile based deferred techniques are also capable of efficient MSAA. Black Rock's microtiled renderer (4x4 tiles) can identify tiles that contain edges, and apply MSAA lighting only to those tiles. It's also very efficient in shadow sampling, since they do not need to sample shadowmaps at all for blocks that are fully lit or fully in shadow (very cost effective if you have complex soft shadows).
I recommend this article for anyone interested in recent high end deferred renderers: http://portal.acm.org/citation.cfm?id=1837046 (Back Rock's "Screen space classification for efficient deferred shading")
Tiled deferred techniques have been used a lot in PS3 games (the Uncharted 1 was one of the first ones to use it). Majority of PS3 deferred renderers seem to be tiled, as tiling suits PS3 architecture better. Tiling is just a bandwidth optimization. The lighting is calculated exactly the same way as in any other deferred renderer, and the g-buffer contents are identical. Also all the same downsides and upsides of other deferred techniques apply.
Tile based deferred techniques are also capable of efficient MSAA. Black Rock's microtiled renderer (4x4 tiles) can identify tiles that contain edges, and apply MSAA lighting only to those tiles. It's also very efficient in shadow sampling, since they do not need to sample shadowmaps at all for blocks that are fully lit or fully in shadow (very cost effective if you have complex soft shadows).
I recommend this article for anyone interested in recent high end deferred renderers: http://portal.acm.org/citation.cfm?id=1837046 (Back Rock's "Screen space classification for efficient deferred shading")
Last edited by a moderator:
