When you select 24-bit, the program operates in 32-bit floating point, which is a useful tool for the program to do all the math without coming up with nasty artifacts. You can dither it all down later, no problem. The 32-bit thing is like an in-house setting for the program to get the job done better.