Originally posted by: Matthias99
Originally posted by: Noob
Originally posted by: keysplayr2003
Originally posted by: Noob
What are the IQ improvements that SM 3.0 offers over 2.0b?
Plases let this not turn into a useless flame war. Let this just be for informational puproses only.
You already know that there is no difference in IQ. I have seen you in enough SM3.0 threads so why are you even asking this? SM3.0 is supposed to offer better performance due to less overhead. SM3.0 is more efficient than SM2.0/2.0b. Nobody has seen any IQ differences as far as I know. In the future, who knows.
I hear something about Parallax Mapping and Tone Mapping or something like that that will be used heavily in U3. People were using that as an example and posting screenshots. So I assume this offers IQ enhancement? Or was that just normal mapping itself. I asked a few questions in the other SM 3.0 thread about some of these new technologies but people were too into the flaming. So I just decided too post my own thread.
Parallax Mapping is a more sophisticated version of normal mapping (it's between regular normal mapping and displacement mapping). I'm not sure if you would need SM3.0 to do it -- it's probably
easier to do with SM3.0, at the very least. Tone mapping is (very basically) a limited form of HDR applied to texture surfaces (at least I think. Maybe I'm confusing it with something else.) I believe you could do Tone Mapping in SM2.0.
PS3.0 basically does not offer any IQ improvements
by itself over PS2.0. What it does is to allow potentially much longer shaders (up to 64K instructions), and it adds dynamic branching instructions (making it easier to write longer and more complex shaders). Now, by using a longer single shader rather than multiple short ones, you might increase performance slightly.
However, today's cards cannot really run exceedingly long shaders without slowing everything down way too much, so this does not do as much as you might think.
VS3.0 adds a few more interesting features. One is geometry instancing, which lets you create multiple copies of an object without having to store multiple copies of it in the video card's memory. This lets you (theoretically) reduce RAM usage and possibly increase performance. However, it is also available via ATI's PS2.0b. The more interesting new feature is that VS3.0 vertex shaders can now access the card's memory; among other things, this allows for hardware displacement mapping, which can do some pretty cool stuff. However, it is likely to have an
enormous performance hit if used in any nontrivial way, since it not only puts a high load on the vertex shaders, but greatly increases the load on both the fixed-function pipeline and pixel shaders (something has to draw those extra triangles).
The
Geforce6 also has support for OpenEXR floating-point framebuffers. This is what is used to do the HDR rendering in Far Cry and Splinter Cell: Chaos Theory. On the GeForce6, AA is disabled while this feature is in use (but theoretically, future cards may not have this restriction). This is
not part of SM3.0, but a totally separate standard that NVIDIA chose to support on NV40. I am not sure if R5XX will support this.
