Which card do you think is gonna power DOOM at E3?

[Nutty]

Both NV30 and R300 have shadow volume in hardware. What NV35 and R350 do (from what I can tell) is allow certain z-functions and stencil (which is how doom3 shadows are done) to be done independently.

What exactly do you mean in hardware tho? I can do hardware shadow volumes on my gf4, just by projecting vertices using a vertex program. Its hardly efficient tho. Some of nv dev rel's mentioned a totally new approach of the hardware calculating proper shadow silhouettes in hardware. This was pie in the sky talk a while back tho.. But NV30 and R300 have no support for accelerating shadow silhouettes AFAIK.

[Lezmaka]

Originally posted by Nutty
What exactly do you mean in hardware tho? I can do hardware shadow volumes on my gf4, just by projecting vertices using a vertex program. Its hardly efficient tho. Some of nv dev rel's mentioned a totally new approach of the hardware calculating proper shadow silhouettes in hardware. This was pie in the sky talk a while back tho.. But NV30 and R300 have no support for accelerating shadow silhouettes AFAIK.

http://www.nvidia.com/docs/lo/2430/SUPP/PO_GFFX_Consumer_030503.pdf

Hardware-acclerated shadow effects with shadow buffers

Now if you still can't get it, I give up.

[Uttar]

You people better trust Nutty, he knows what he's talking about :)

Shadow Buffers *isn't* the same thing as Shadow Volumes.
Nutty is pretty much right by saying there is little hardware optimization going on for shadow volumes.


Uttar

[Uttar]

BTW, it seems shadows will be a more important part of the NV35 demos - as could have been supposed considering there's a Shadow Volume Accelerator. In fact, I think they'll even show a few nifty things with shadows. But don't worry, there's less than 2 weeks left before it's all shown publicly :)

Oh, and BTW, the demos still aren't optimized. Good luck to all the nVidia employees who're gonna get in big trouble if they run too slowly at launch!


Uttar

[DSC]

Does this also mean that the drivers are not optimised as much for shadow buffer effects? Or just the demos for NV35? :confused:

[Nutty]

Shadow buffer implementations use pretty much generic features. You basically just render the Z buffer of a scene from the lights point of view straight to a texture, and use this to calculate which fragments are shadowed and which aren't.

Basically it just uses the render to texture features, which have been optimized quite a bit recently. Just its a Z buffer format texture rather than RGB(A).

To get this fast, you need fast Z writing. (This is why Nv30 does 8x1 in Z write only. i.e. it speeds shadow buffers up alot. As there is no texturing going on at this point). And decent render to texture speed.