I could get flamed for stating the obvious here, but in my Gentoo 64 installation this was giving me the screaming heebies until I fixed it 5 minutes ago.

No matter what I did, emerge out the wazoo, whatever, I couldn't get OpenGL working on my GF5200 card and Xorg. Glxinfo complained about not being able to find RGB, and so on. I finally found the problem (in /var/log/Xorg.0.log).

NVidia OpenGL apparently only works in 16 or 24-bit. When my KDE session started up, it would change resolutions on me before settling down, and nothing GL-based would work. The reason is that no matter what your "Display" Depth is set to in the Screen section of Xorg.conf, it needs a DefaultDepth entry of 16 or 24 for OpenGL to load properly. Mine looks like this:

Section "Screen"
Identifier "Screen0"
Device "NVidiaCard"
Monitor "Monitor0"
DefaultDepth 24
SubSection "Display"
Viewport 0 0
Depth 24
Modes "1600x1200"
EndSubSection
EndSection

Hope this helps someone.

Damien

That is true. You see, 32bit color isn't really 32bit color, it is 24bit color with an 8 bit Z-buffer (if you don't believe me then load up windows sometime and look at the actual number of colors in 24 and 32 bit mode... yes they are the same). It is my understanding that the nvidia driver assumes the buffer so it uses a max of 24 bit color.

Ok... I need to clarify... some graphics card people use the extra 8 bits in "32-bit" color for a Z-Buffer (or sometimes other buffers). I believe nVidia does this, but I'm not totally sure.

The biggest reason for 32bit color originally was so that the data would always by aligned in a single word. For example... if you have 24bit color (24 bit color = true color = 16,777,216 colors) then three colors held in memory back to back would look something like this:

0x00 11 11 11 22
0x01 22 22 33 33
0x02 33 xx xx xx

I hope this wasn't too confusing.... but the whole point is the 2nd and 3rd colors overlap the word boundries and thus it is much more difficult to get access to the data on a machine with 32bit words (you need both words in memory instead of just one.. which means you end up loading 64bits of data to access 24bits in a lot of cases).

In 32 bit color you insure each one gets a full word, and you never have to deal with things crossing word boundries. This trick also works for 64 bit machines because the size of a color is evenly divisible by the word size.

My understanding from nVidia Tech support is that they byte align there colors by default so all you are setting is the actual number when you set the display mode.