Very nice review, especially by also putting the 9800 in concerning instancing.
Something interesting I noticed... The new SM2.0b and SM3.0 mostly seem to increase performance, when the GPU was having an easy task allready, but not when the going gets tough...
To explain what I mean, just look at the first graph showing the results of SM2.0b.
The regulator and training scenes perform at 59 and 61 fps resp. using 2.0. Going to 2.0b gives only
1 fps gain.
Now, the Research and Volcano perform at much higher framerates, 74 and 83 resp. Here, going to SM2.0b gives gains of 7 and 12 gain.
Everyone is excited at those 12fps gain going from 83 tot 95 fps...
But actually, that's a bit useless, as it was allready 20 fps faster than the other scenes.... It's the other scenes where you'd want a boost. Because those are limiting the settings at which you play. (Unless you're switching settings between scenes...)
Now, I don't want to sound negative, but looking at the new pixel shaders from this angle, means that in most cases on both ATI and NVidia you gain about 1 fps. (The performance gain of the lowest performing scene)
Only at 4xAA does the 6800GT show a little bigger improvement up to 3 fps...
Ofcourse, these benches don't neccesarily tell the whole story, and any performance increase is always good, but the improvements do seem a bit 'acadamic'. Not only for instancing, but also for the pixel shaders.
Before someone acuses me of being negative... This ofcourse means that Crytek did a fantastic job programming Facry. It's so well optimized, that even improved techniques from hardware that didn't exist when the game was released, can't improve the performance in a usefull way.
