BRiT said:
More bits = more precision = less quirkiness on games which use truely expansive fields of view.
field of view doesn't get involved here, it's almost exclusively about the
near clipping distance; you could have as large FOV as you like with any clipping distance*.
apparently derek's problem originates from the fact that in his game universe the player can look from the POV of very small and very large objects (relatively to each other, that is) alike, e.g. from the cockpit of a battlecruiser, just as well as from the visor of a helmet. problem is, in the case of looking from the helmet's visor, visible objects may come much closer to your POV than in the battlecruiser's cockpit case (cases taken realtively to each other). now, this really matters if you want to preserve
the same universe resolution as seen from the marine's and from the battlecruiser's viewpoints alike, i.e. you indeed have to move the near clipping distance closer to the POV for the marine case.
of course, precision of the z-buffer is something people have been tackling with since the very invention of the beast. the final 'efficiency' of the z-buffer (i.e. how well it serves its occlussion purpose) is a function of just two arguments: the raw precision of the buffer (i.e. it having n-bits per entry), and the near and far clipping ranges of the (perspective**) projection transform of your pipeline (although the near clipping distance is the far more important among those two). usually when none of the two factores can be effectively controlled (i.e. when you can neither set higher z-buffer bitness nor push the near clipping distance further ahead) some tricks come into play - most commonly said tricks come down to some depth-partitioning scheme which would allow us to divide our scene into depth "layers", and render those front-to-back, each time clearing the z-buffer. this would provide each of those layers with the full 'precision' of the z-buffer. of course, such a trick costs performance, as it can be though as imposing extra 'passes' to render the scene.
bottomline of it all being, striving for higher bitness of the z-buffer is good, but no matter how many z-bits the lates-'n'-greates hardware has, there always will be large-enough universes populated with small and large 'characters' of sufficient 'disproportions' to make that z-buffer bitness look bad.
* FOV - in 2d (for clarity), the angle formed from looking at a certain view-span from a certain view-distance.
** of course, it only matters iff the projection transform is
perspective, as with a
parallel projection z-buffer would be completely equi-precise across its whole range.