Originally Posted by Chewi
The two fundamental modes that you can use are
1) multi-head/single-GPU - called Zaphod (but increasingly incorrect with the AMD >2 display cards coming out).
This mode leaves you with two X heads, :0.0 and :0.1 which are more or less independent universes. With fglrx, you can do "aticonfig --initial=dual-head" and it should just light up.
2) RANDR 1.2
This mode gives you two viewports over a single surface and the ability to size and position independently. Benefit is configurability, disadvantage is application support. If you are using a RANDR 1.3 capable distribution, you can also set "--primary" to provide a hint to gnome-panel et al about which monitor is your "center of the universe".
Disadvantage here is that applications are mostly naive when it comes to multi-monitor modes. They may listen to the xinerama extension and place things intelligently, they may not. This goes right up and down the stack.
With fglrx, the only caveat here is that you need to set the virtual size to the max bounding square of a desktop that you care about. Then layouts should work.
If you go with this mode, you are likely to get quirks with apps, and to resolve those quirks you probably need to work upstream with the apps to get it going how you would like it.