it's a laptop, and runs an intel hd4000 card by default, maybe the ATI can be switched on only on fullscreen mode. then I get 29 fps on 1024x768. Slin's code results in a further speed increase to 32 fps. I mean pp. the obj is about 12 now for both.

I used 4x16bits floating point textures to render the world position and a 1x16bits floating point texture for the depth maps. Maybe you can change them to 32 bits images (14444 and 14) and tell us how it went

But that is not the real clue...

Both render pipelines do same computations, but deferred one does once per screen pixel instead of once per polygon pixel.

You can complicate or simplify both pipelines that you will get nearly same proportional performance. I did not look into Slins PCF technique, but in native pssm shadows there is a faster one. I simply choose the slower one to emphatize the deferring technique benefits.

What I learned of this technique is that writing into a 4x32bits bitmap for the deferred lightening takes its time, and it has to be smaller than the time taken to direct render the overlapped polygons. If you render an nearly empty scenery, deferred techniques are slower.

I just made a larger level with clip_far 8000 and with 2048 pssm resolution from own resources (large terrain with trees having lod), 2500 trees on a 20000x20000 area, everything looks fine, fps can go up sometimes to 39.
as I see it should use an own terrain shader, and the sky cube is also messed up. I hope you'll make additional stuff

did you use the native pssm shadows with the slow pcf technique on same level?

---
Yes, all the objects in the scene should consider that the rendering is deferred.


edited_____
I forgot to mention that both object materials and depth shader have a double sided rendering

the pcf function I posted is the same as yours, just using tex2Dlod instead of tex2D as the difference does not matter for the depthmap sampling, but tex2Dlod allows the graphics card to use dynamic branching, which in this case means a huge speed benefit, as without the PCF sampling is done for each depthmap for every pixel instead of just the one it needs to.

thank you for the info!

i should read more about samplers
but in native pssm sahdows...

Hey, it was dark, because im using a different color space profile..
But i forgot one thing, "to turn of the powerhybrid technology" (modern stuff..) now , after putting the CPU on the normal performance.. i get 20FPS , but i tweaked the script to lower the tree density : now i dont know the original value :S BUT the shadoow artificat its very notorious.

Im using a laptop Core i7, GT650M :S

Shadermodel 5.0, Cuda, and physX :S
http://www.tiikoni.com/tis/view/?id=9f10b44