To clarify this, I meant DX9-level stuff that is instanced_arrays, draw_instanced, half_float_pixel (this one is in Catalyst 8.5 only) and maybe rgtc or at least the 1-channel compressed texture format known as ATI1N in the DX9 world (the ATI_*_3dc extension gives us just the 2-channel one).
What’s the point in exposing EXT_depth_buffer_float when there is an ARB equivalent standardized?
I got all of them on the 4850 (but I’m using newer 8.54 drivers, didn’t try the official cat8.9) - however, they didn’t show up at first (firing up glview right after installation+reboot) - seems like I had to do some gl rendering or something to active them initially, but now they’re reported (haven’t tried them yet though). So, try again, maybe run the test and see if they show up.
Is this still the case? Do you need to do some rendering in order for the correct extensions to be reported?
What do you mean - “you need to do some rendering in order for the correct extensions to be reported”?
I am using ARB_depth_buffer_float extension on HD2400 with catalyst 8.12 (Vista32). It reports itself in GL_EXTENSIONS string and GL_DEPTH_COMPONENT32F texture format works correctly with my shadow mapping (instead of GL_DEPTH_COMPONENT32).
Although I get no Z-buffer at all (everything render like z-test is disabled) when I try to use GL_DEPTH32F_STENCIL8 format for depth&stencil texture used as attachment for FBO. But hey - it’s ATI 
I’m glad that at least GL_DEPTH_COMPONENT32F works fine.