Here is a plug in I recently purchased.  It offer different dithers and you can actually see where you are getting distortion from common dithering in real-time.  It's a bit pricey but it is also pretty amazing.
 
http://www.sonnoxplugins.com/pub/plugins/products/pro-codec.htm
 
 
Kevin

I noticed Brandon recommends Powr3 in the Cake vid's.. he feels it sounds better.  I don't hear a difference in Triangle, Powr3 or when I let Soundforge do it.  All personal preference It seems  ; )

If you want to clearly hear the differences between the various dither algorithms, do the following:

Take a single snare drum hit and run that thru a reverb with an extended tail (bounce it down).
Now listen to just the decay of the reverb on a good set of headphones.
Open up the headphone amp so you can hear the decay all the way to the noise-floor.
NOTE:  Be careful when doing this!

Now... apply each dither algorithm and bounce (each) down to a separate 16Bit audio file.
Using the procedure above (headphones with amp up loud)... listen the resultant 16Bit audio files.
You'll find that the reverb decays are "smoother" on some than others.
You'll also find that some dither algorithms are more audible than others.

After doing this test, I find Pow-r algorithm 1 to be the overall best in Sonar's offerings.
Smooth results... and the dither isn't too audible (comparatively speaking).
The dither noise in Pow-r algorithm 3 (noise-shaped) "irritates" my ears.

In regards to smooth sounding results (and dither noise relatively quiet) , I find Pow-r algorithm 1 and UV22-HR right about equal.

Before anyone says it (Hi Bit  ), this type of listening *does* magnify the results (not a typical listening scenario).
That said, if I know one algorithm produces smoother less-audible results (even though the difference may be extremely subtle), I'm going with that algorithm.

Interesting experiment. I've just done this will all 5 algorithms, and I must say Jim, after doing this I think I prefer Pow-r 1 as well.

The audible difference between the others isn't huge, though it's easy enough to eliminate Rectangular and Pow-r 3.

What is interesting is if you zoom in on the resultant waveforms during the fade, and here you can see less than smooth results with both of those (Rect / Powr3)

Powr3 has a bigger wave than the others, triangular is the smallest, Rectangular is chopped/bitty, and powr2 isn't too bad, although again, Pow-r 1 seems to stand out as the best in terms of smooth fade and overall loudness.

I've been using Powr-3 for my final export mix, since it seems to be "best" according to Sonar documentation.  Jim & Bristol, you have me tempted to switch to Powr-1.  I don't doubt the results of your experiment, but that's only 1 scenario, so I wonder if Powr-1 would be superior in other scenarios as well ... from a dense rock mix to a quiet orchestral mix to a pristine solo piano ...

And no, I don't plan to do the testing.  It would be great to hear from someone at Cakewalk on this topic.

Interesting experiment. I've just done this will all 5 algorithms, and I must say Jim, after doing this I think I prefer Pow-r 1 as well.

 
Yeah, Pow-r alg. 1 was clearly superior (in this scenario) to my ears.
In my way of thinking, the "decays into the noise-floor" is the only time you'd year quantization noise.
(You're not going to hear it while music is playing at normal levels)
So I think the test scenario is pretty valid...
 
Note:  Your ears may not hear things exactly the same.  This type of test is pretty quick/easy to setup/execute.  Within a few minutes, you'll know which alrgorithm you prefer... and you can decide for yourself if the difference is appreciable. 

I think I'm going to repeat the experiment.
I set this up on my secondary rig, which is in a room with a lot of ambient noise and only limited headphone volume.

When I get a chance I'll do it again on my main rig

But if we give our mix to a Pro (say Alan Parsons) what would they say if we chose the wrong dither?

Of course if you're giving your mix to anyone - you wouldn't dither at all
 
*They* would dither when they're done with it...

Jim Roseberry

Interesting experiment. I've just done this will all 5 algorithms, and I must say Jim, after doing this I think I prefer Pow-r 1 as well.

 
Yeah, Pow-r alg. 1 was clearly superior (in this scenario) to my ears.
In my way of thinking, the "decays into the noise-floor" is the only time you'd year quantization noise.
(You're not going to hear it while music is playing at normal levels)
So I think the test scenario is pretty valid...
 
Note:  You're ears may not hear things exactly the same.  This type of test is pretty quick/easy to setup/execute.  Within a few minutes, you'll know which alrgorithm you prefer... and you can decide for yourself if the difference is appreciable. 

+1 to this test.

It reveals why it should be done whenever you are reducing bit depth, the benefit becomes clear.

Once you've heard the effects of not applying it they become easier to spot and choose the appropitate algo for your needs. I've yet to come across something where 'Triangular' hasn't been sufficient for me but I'm thinking now I might re-visit Pow-R 1.

I'm sure Pow-R 3 has more advocates on account of it sounds like it must be better than 1 because it's 3 right?... i.e.  My mixes haz 2 more Pow-R's than anyone elses...

jbraner

But if we give our mix to a Pro (say Alan Parsons) what would they say if we chose the wrong dither?

Of course if you're giving your mix to anyone - you wouldn't dither at all
 
*They* would dither when they're done with it...

Unless of course you are supplying a 16 bit version of your mix, then it's your job.

*They* can just treat it like a native 16 bit file then, if that is what has been specified for delivery.

Don't forget here either that the effects of dithering are cumulative so only do it once as the last thing you do at the point of reducing bit depth.