I did a search on the bit depth for playback, but the difference between 32bit fixedpoint and 24bit fixedpoint padded to 32bit has not been explained yet. if i'm using a card that only supports 24bits won't it truncate the audio down to 24bits? also, when the 24bit audio stream is padded to 32bits what is the padding? just filler? zero's?

If your card supports 24bits than use 24bits padded to 32bits, this is done for efficiency reasons (I assume that for all cards it'll be more efficient to send 32bits; which isn't correct). But you should try posting your card's specs here.

i've got an m-audio revolution 7.1 which is a 24bit card and supports upto 192khz output. input is only 96khz.

I was wondering this myself. I just bought the Prodigy 7.1, which is also a 24bit card.

i think it uses the same chipset, does it not? the Via Envy24ht aka ice1724.

On my DMX 6fire (and possibly some other Envy24(HT) cards) it is prefered to use 24-bit padded to 32-bit because DACs on the card are 24-bit, while mixing is done at 32-bit, so it is more efficient.

Terratect explains this in manual this way:
1) if you use 24-bit, you have "glory to the CPU mode" (but card doesn't downsample for DACs)
2) if you use 32-bit, you have "happy CPU mode" (but card downsamples)

To sum all this - 24-bit fixed point padded to 32-bit is the best choice.

What about a SBLive?

16-bit fixed

Dither is up to you.