Mixdown 16-bit or 32?

When you mixdown to master, do you choose 16-bit or 32

Could someone explain the big difference between them, I know 32-bit has more quality but can you really tell them apart?

Thanks

32-bit mixing prevents the loss of resolution you get if you alter the amplitude of a file - and that means virtually any form of modification at all to it - filtering, dynamics processing, adding effects, the lot. With a 16-bit file the resolution loss you'd get from, say, attenuating a file by 12dB in a mix would be to reduce the contribution of that part of the sound to effectively a 14-bit one. You don't lose the resolution at all if you do the mix as 32-bit floating point.

But what actually happens in AA2.0 is that the mix will be done in 32-bit FP anyway, and it's just the results that will be stored as 16-bit if you select this resolution. The only thing you can alter within a session as far as resolution is concerned is the bounce-down resolution, which can either be 16 or 32-bit. If you are going to master a mixdown after the event, then the mix should be 32-bit, regardless. If you are going to write direct to a CD with no other intervention, then mixing down to 16-bit is fine in theory, but you should also save a 32-bit copy, just in case...

Can you hear the difference? On a complex mix, most certainly.

Thanks for the enlightment, and 32-bit it is!