Bristol_Jonesey
64 bit is of course even more accurate and I always engage this option.

By this point it's only more accurate in theory as it's almost impossible for 32bit floating point not to preserve everything actually useful.

If you are sending your mix out for mastering, why not just check the mastering studio for what format/resolution they recommend? Generally, if you are entrusting your files to a specialist, he will probably not want them altered from the format used when mixing. Although there is some processing done in mastering, work is relatively limited and typically on only two tracks, so the chance of significant rounding errors is pretty small and a massive pile of bits is probably not going to add much in the way of insurance. It certainly does not make sense to send out a file that is rendered at higher resolution than used in mixing. 

I've never been asked for a 32 bit file, across work done for record labels, video games and TV.
 
Typically, as things stand right now, mastering engineers will ask for a 24 bit file.
 
It's important not to confuse the internal processing bit depth with the bit depth going on at the master bus output. That's dependent on your project bit depth, and if you export at the project bit depth, then you are exporting what you've been listening to as you've been working.
 

I agree with Johns x 2.  I have never been asked for nor had cause to export @ 32bit.  I understand the theory behind it but am sceptical of any real world advantage beyond confirmational bias. Still - if it makes you happy:) 

slartabartfast
If you are sending your mix out for mastering, why not just check the mastering studio for what format/resolution they recommend?

^^^^ this

Ah, conflicting opinions.
I am "on the fence" with this. The audible difference is very slight.
For now, I am mixing down to 32 bit (my understanding is that the mix process dithers on a 24 bit export from its internal 32 bit processing state?).

pdarg
Ah, conflicting opinions.
I am "on the fence" with this. The audible difference is very slight.

 
The only way you will get an audible difference is if you are either clipping or are boosting the level by tens of dB's after export. 
 

my understanding is that the mix process dithers on a 24 bit export from its internal 32 bit processing state?

If you have dither enabled on export, then it dithers. But dither at 24bit levels is not remotely close to being audible (unless you are boosting the level by tens of dB's after export).

If you hear an audible difference, the issue is probably what you are listening to the wav files in. If you use Winamp or windows media player or the like it can have issues with playing floating point files incorrectly because of how it interprets them. In something like Wavelab or Sound Forge they should sound the same, as your soundcard can only handle 24-bit output anyhow (and realistically resolves only maybe 20-21 bits).
 
In terms of the actual difference, both files have 24-bits of precision, a 32-bit FP file just maintains that over a large range of volumes. So if you take a 24-bit file, and cut the volume 24dB, you then only have 20-bits of precision in that signal since the 4 most significant bits are now silence. With a 32-bit file, the 24-bits of precision is maintained down to a very low level. Likewise a floating point signal can exceed 0dBFS whereas an integer one can't and will clip.
 
However no matter what, it is getting converted down when it gets played back. So all the extra precision and so on doesn't matter during playback. Rather it is for processing. Any time you round something down when doing math you are introducing quantization error so you want to round only on the last step of an operation. Hence why DAWs n' plugins do everything in floating point.

Okay - I agree.
So, since there is at least one last step of processing of the WAV mix file (after it is exported from Sonar), it should remain in 32 bit until it is put into its final format (in this case, an MP3 file).

It isn't important that it remain 32-bit, but yes I'd leave it as 32-bit until you are done mastering. I don't send my stuff off for professional mastering, since I'm just toying around, but my process is to export to 32-bit float from Sonar and load that in to Sound Forge for mastering. Once I'm all done, the last step is to dither down to 24-bit. I then convert that to MP3, FLAC, or whatever I'm going to have it as finally. It is not necessary to do it that way, but it is just additional insurance against any quantization error.
 
Same reason I have Sonar run using its 64-bit mixing engine. It isn't necessary, and I'm betting most files would not just audibly null but bit-null if you compared 32-bit vs 64-bit, however the extra precision doesn't cost me anything I don't have plenty of (CPU and memory) and it is insurance against any kind of quantization (32-bit is 24-bit precision, 8-bit range, 64-bit is 53-bit precision 11-bit range).