+ rendered in 1280x768 @ 60fps (the arcade panel is 1366x768 so there are black bars)
+ The development had begun before NVIDIA provided OpenGL perf analyzer, so they developed an
in-house tool to maintain 60fps. (vertex in the picture includes that of self-shadow) The bottleneck was not in the AGP bus but in the pixel load.
+ OpenGL2.0 w/ ARB assembler for shader programming. NVIDIA extensions are employed too. (Not all LINDBERGH games are like this, HOD4 is GLSL, VT3 is Cg. So they apparently had some difficulties to transfer VF5's shader code to PS3, but in the case of VT3 it was easier)
+ 1 character = 40K polys, background = 100K - 300K (VF4: 12K and 50K)
+ Basic shadow is done by depth shadow (1024x1024 map) with a screen-coordinate-based mask for anti-aliasing against shadows, though this method can't hide shadow aliasing in up-close shots. NVIDIA shadow (Hardware Percentage Closer Filter) is employed. Adaptive multisampling is a bit too expensive to maintain 60fps because of load fluctuation. VF4 was stencil shadow volume.
+ Shadows on the ground are silhouette projective texture mapping. Far backgrounds have prebaked shadows. Some close backgrounds such as columns in buildings have real shadows generated with depth shadow.
+ Skin, hair, clothes are made with diffuse, specular, and normal maps. Some parts such as Akira's wear has special shaders. Pai's dress is normal mapped with an environmental map mask with diffuse and specular adjusted.
+ Skin has no special shaders such as SSS. Like other parts it is made by adjusting parameters in basic shaders. Female characters have special lighting to reduce shadows.
+ The backlight expression is simple blending of the edge color and the light color when the light comes out of edges. Some irregular cases are corrected by adjusting parameters.
+ There are 4 versions of Dural and 2 among them have special shaders. The one has diffraction simulation by vertex shaders to procedurally generate the rainbow coloring. The other is the refraction version of the first one.
+ VTF (Vertex Texture Fetching) in the NVIDIA GPU is used very widely for VF5. Apparently this function is essential for VF5 since it's very preferable that GPU can do all basic geometry.
+ For water wave simulation, VS reads a FP32 texture height map by VTF, then does displacement mapping to transform vertices on the water. With this FP32 wave height map converted into a normal map, PS does shadow processing. Muddy water is drawn by PS using velocity/power information in FP32 wave height map. Thanks to VTF, VS can detect the collision of water splashes and it's used to generate ripples when each water splash hits the water.
+ The snowfield also uses VTF. Shadows in the snow show pseudo-subsurface-scattering by the texture-space blur.
+ Fog is also produced by VTF. Fog density is represented by a FP32 texture. So fog reacts to characters' movement.
+ The render target buffer is FP16-64. Textures in environmental maps and reflection maps are in the ARGB integer format + luminance information packed in alpha when required. Apparently FP16-64 texture fetch is slow because of filtering and not suitable for 60fps right now.
+ VTF appears in the tone mapping process too. Making an HDR frame into a 1x1 pixel texture, VS reads it by VTF and the vertex pipeline calculates average luminance.
+ An in-house engine does physics simulation. It only processes local collision such as hair and clothes. Character-to-character collision is not suitable for simulation because of the nature of the VF franchise. (VF3 did Inverse Kinematics but it was removed in VF4)
