Noob questions about some concepts

[Carnage1986]

Hi. I have some questions about image quality and fillrate? First of all i saw two charts and they attracted my attention:

Here's my questions:

1/What is FP16, FP32, INT8, fp10 texels? What is the relevance of floating points and integers with textures. Is it simply bit depth of textures?

2/Why some scenes need more than one texels? (For example 4xFP16, 2xINT8)

3/Why fillrate drops on some of AMD cards when FP16 texels used, and why some others suffer performance loss on same scenario? Because of bandwith limits or lack of ROPs?

4/Wikipedia says "complexity of scene affects actual fillrate". Why?

 

[Anarchist420]

3/Why fillrate drops on some of AMD cards when FP16 texels used, and why some others suffer performance loss on same scenario? Because of bandwith limits or lack of ROPs?

Because the ROPs do 64 bit formats at half the speed they do 32 bit formats. theoretical fillrate is independent of bandwidth.

2/Why some scenes need more than one texels? (For example 4xFP16, 2xINT8)

each texel has 3 or 4 components, which are usually red, green, blue, and alpha.

1/What is FP16, FP32, INT8, fp10 texels? What is the relevance of floating points and integers with textures. Is it simply bit depth of textures?

floating point textures allow more fractional values with fewer integer values per bit; integer values have more unique values but less fractional bits.

4/Wikipedia says "complexity of scene affects actual fillrate". Why?

more AA (except for post processing AA) reduces fillrate, more render targets reduce fill rate, and higher resolution reduces fillrate.