I know NVIDIA cards can perform Percentage Closer Filtering on hardware, is it just as simple as applying GL_LINEAR to ARB_SHADOW texture? Is it all?

It's just bi-linear filtering of the four nearest shadowmap comparison results. It usually looks OK, but it's absolutely unrelated to PCF. PCF is something entirely different and involves taking multiple comparisons per texel (usually 16 or more in a 4x4 pattern).

Quote:Original post by Anonymous Poster
It's just bi-linear filtering of the four nearest shadowmap comparison results. It usually looks OK, but it's absolutely unrelated to PCF. PCF is something entirely different and involves taking multiple comparisons per texel (usually 16 or more in a 4x4 pattern).

How is it entirely different? Bilinear filtering takes into account multiple comparisons(4).

Nvidias bilinear method takes one sample per texel, and mixes the results together. PCF takes multiple samples per texel, and operates on subtexel level. You can think of PCF as a kind of multisampling for shadowmaps.

Compare it to a colour image antialias algorithm: NVidias method would be like using an un-antialiased image and removing the jaggies by simply blurring it. PCF would be like rendering the image at a much larger resolution, and sampling it down (supersampling).

Since PCF processes much more information per texel, its quality is generally a lot better than NVidias hack. But it also takes alot more performance if using many samples. Eventually the quality will be limited by the resolution of the shadowmap itself.