robbyk 
And secondly, I knew dithering is adding noise, but I didn't think it could add up to a half of dB in a 3 minute song.
 

It should only add a much smaller fraction of a dB than .5dB (peak) at 16 bit levels.
 
It's common for intersample peaks (at 44.1 or 48kHz) to be in the +1 to ~+3dB range. It's possible to get higher, but the higher they go it becomes much less likely that you'll see them in the real world.
 
And it's my impression that most good digital-lookahead-peak-limiter/loudness-maximizer thingies include oversampling, both for intersample peak detection and for cleaner processing. So if you're going to use such a thing, the oversampling options should always be turned on.

drewfx1

robbyk 
And secondly, I knew dithering is adding noise, but I didn't think it could add up to a half of dB in a 3 minute song.
 

It should only add a much smaller fraction of a dB than .5dB (peak) at 16 bit levels.

drewfx1
 
Very cool explanation. Yes my range of increase has been up to 0.5 dB. You are right.
 
Now I think it's time to end my eMusician subscription (getting less and less applicable to my interests) and begin the arduous process of reading and rereading the wonderful Izotope guides for my evening wind down.
 
You all were a very big help on this!

...I had thought Ozone was a better dithering plugin than others...

iZotope says it is, natch. It may even be true. But the difference between the world's best dither algorithm and the crappiest one isn't huge, nor is it hugely important. I would not choose my mastering limiter based on the quality of its dither.
 
Having said that, I've long used MBIT+ for dither. But that's mainly just for convenience because it's right there in the multi-purpose plugin I'm already using.
 
Every aspect of iZotope products I've ever examined closely has impressed me with its quality, e.g. zero aliasing from any version of their harmonic exciters (and the only such product I tested that could claim that). It's therefore safe to assume they know what they're doing with dither, too. But even if I were to discover that a better algorithm exists I'd probably continue using MBIT+ because dither just doesn't really matter that much.

bitflipper
iZotope says it is, natch. It may even be true. But the difference between the world's best dither algorithm and the crappiest one isn't huge, nor is it hugely important.
 

That is a very good point, probably a matter of degrees :)
 
I'm sure I also will stick with Mbit+, high, 16, normal, DC filter on as I have since the early days of Ozone 4.