Interesting .. yeah the parallax mapping on the wall to the right - not the bricks- looks a bit streched out .
Grrr the more i see .. the more i want the demo
! the problem is with my slow connection i cant get it via torrent
anyway maybe we should post Demirug's explenation about why SM2.0 wasen't used in SC3 , it looks fair to me but maybe some people might have more info on that or dont agree with it.
Originally Posted by Demirug
2.B designed for this matter? Who have told you this? Even 2.A have a stronger featureset but the NV3X cards with the poor FP performance make it hardly usefull at all. But this is an other story.
ubi uses something that is called "übershader". At the moment if you have an DirectX engine this concept can only used meaningfull with SM3. SM2.0 did not have sufficient instruction slots. SM2.A and SM2.B did provide enough slots but missing branching. This force the chip to execute to many instructions for every single pixel. A SM3 chip will step over this unused instructions. I am not talking about dynamic branching as we allready know that a NV4X need large pixelblocks to make this run well. I am talking about static branching.
The next problem with some of the shaders is that the vertexshader need to send 10 vectors to the pixelshader. SM2 (all variants) only supports 8 vectors. It is possible to reduce this number because many of this vector values depend on the position. In this case you can only send the position but than you have to do all the calculations in the pixelshader. This will slow you down again.
You can work around this problems if you use an other concept but this means that you have to write extension for the engine and the toolchain you allready have.