Anisotropic Filtering with CATALYST

[Ozo8500]

I've never tested anisotropic filtering before, but the results I get with this new driver are not very impressive.

If I use trilinear filtering with OpenGL (i.e. min filter = GL_LINEAR_MIPMAP_LINEAR) with NO anisotropic filtering, I get quite good results. Of course, I get the filtering between texture LODs as expected.

However, when I go in the control panel to enable anisotropic filtering (2x up to 16x), the Radeon does not filter between mipmaps anymore! What I get is a better filtering within a given mipmap, but no filtering between mipmaps (i.e. sharp transitions between mipmaps) It looks more like GL_LINEAR_MIPMAP_NEAREST filtering than anything else. I

Does anyone know if it is a known limitation of the Radeon 8500 chip? Do we know if other 3D chips (like nVidia) work the same?


Ozo8500

P.S.: You can easily test it yourself by compiling this simple opengl program: http://www.opengl.org/developers/code/examples/redbook/mipmap.c
Don't forget to change the filtering modes to GL_LINEAR_MIPMAP_LINEAR instead of GL_NEAREST_MIPMAP_NEAREST (for minification) and to GL_LINEAR instead of GL_NEAREST (for magnification).

 

[Camster]

The 8500 can do either trilinear or Aniso but not both together. Leave it set to Bilinear and set Aniso to 16x for best look/performance I would say.

 

[Wunderchu]

yes, Camster is correct,


the Radeon 8500 does not support anisotropic fitering with trilinear filtering at the same time



... like Camster says, the best look you can get with the Radeon 8500 , is probaby with max anisotropic filtering with bilinear filtering at the same time



P.S. I would recommend 16X anisotropic filtering for almost all your games, because anisotropic filtering on the Radeon 8500 does not effect performance too much ...

 

[Ozo8500]

... like Camster says, the best look you can get with the Radeon 8500 , is probaby with max anisotropic filtering with bilinear filtering at the same time

Hum ... In many cases, that would probably produce this sharp (and annoying) mipmap transition line. Of course, the line is pushed back because of anisotropic filtering, but it is still there.

For sure, the more different mipmaps are, the more obvious this transition line will be. And since our brain is very good at picking artifacts ...

I guess it is a tradeoff: we have to choose between more sharpness (anisotropic filtering) or more smoothness (trilinear filtering).

Ozo8500.