Thanks to everyone for shedding some knowledge. 

bitflipper
No, the only downside to 64-bit processing is that it places greater strain on system resources: more RAM used, more CPU cycles, more disk storage.
 
Dither is irrelevant in either 32- or 64-bit processing. There is some debate over whether it is relevant to 24-bit integers, but a strong case can be made that it is also a non-factor in that scenario. Yes, your 64-bit files do need to be truncated to 24 bits, but the resolution is high enough that no significant precision is lost. Remember, all this is happening at ridiculously quiet levels way below the level of audibility.

True the difference is very subtle. Two factors might have made it quite evident in my export. 

1. I had bounced the strings from another project and imported the strings track to the current project. I might have added dithering during the first export from there as well. So successive dithering might have made my case obvious on the strings.

2. I did terrible sound staging prior to the mix and ended up with a mix with master bus peaking at abt -12dbfs. Not willing to restage the already good sounding mix I simply kept the master fader at +6. I guess by doing this I might have brought up the noise level and this might have somehow affected my track when I bounced at 24bit with no dithering. 

Noel Borthwick [Cakewalk]
Generally dithering has an effect when you are exporting to integer formats (16/24 bit).
If you intend to use it for mastering you should export to 32 or 64 bit float with dithering turned off.

Thanks Noel. I understand the daw is internally handling audio in 32bit or 64bit. ButIhave kept my record, render and import bit depth to 24. So now if I export my mix for mastering in 32 or 64bit format with dithering turned off, will it still benefit?
If thats possible it will be of great use, I can totally forget abt dithering and make all my exports 32bit.

michael diemer

bitflipper

michael diemer
Wow, this stuff is so far above my head. What about music for full orchestra? Should dithering be used there? I usually just go with the default setting. Should I change it?

No. The violins will supply all the dithering noise you need.

Yeah, you got that right! The low strings also do a great job muddying up the mids, too. Strings remain the most challenging thing to get right. but then I have heard actual performances where the high violins sound terrible. Just too many instruments, no way they can be in tune and in time with each other.
 
So, I guess I should turn dithering off altogether? 

Whats interesting is I heard Yoad Nevo telling he would keep the dithering off since the waves plugins he used already emulate some analog noise in the background. 
https://www.youtube.com/watch?v=2V9d6zQzcrg

bvideo
These discussions on dithering and hearing the difference are interesting. For dithering meant to apply to export, there is an easily repeatable experiment that can be done. The OP would be in a very good position to try this.
 
Export multiple files, each time using a different dither setting. Twice for example: once with no dithering and once with a dithering option that sounds different. Then in a bare project, put those two files, one out of phase. Based on my understanding of the definition of dither, those files should cancel out down to the least significant bit.
 
The resulting file should be noise, and it should be very hard to hear when played back at the same volume that would normally be used to listen to the exported original. Certainly if the original file does not use the dynamic range available in the original format (e.g. 24 bits), the dither becomes a larger proportion of the signal, and may cause audible artifacts.
 
In any case, if samples in the resulting file are larger than '1' (literally the value 1 in a possible range of -838608 to +8388607) there could be a bug in the dither code or an error in the application of it. (Edit: there are some controls on the timeline to zoom vertically on the waveform to see 1's and easily see samples out of that range)
 
Does the canceled output sound like noise or does it have any pitch or audible pattern (turn up the volume to hear it)? The noise can be subjected to frequency analysis to see if the spectrum looks like noise in the frequency range predicted by the description of the dither option.
 

Will try that.

First, there is little benefit to rendering at 24 bits over 32. Disk files are 25% smaller, that's all. There's no audible downside to running at 32 bits throughout. Even if you're importing 24-bit audio or using 24-bit samples, all CbB is going to do is tack on zeroes to pad them out to 32 bits, which doesn't change the sound at all. However, the 32-bit files will be far less susceptible to problems resulting from subsequent modifications to the data. 
 
If you export your 24-bit project to 32 bits, the same thing happens: zeroes are appended to each data word. Dither isn't necessary whether exporting to either 24 or 32. Only if you export to 16 bits do you need dither. Even then, you might be surprised to find that it makes no difference there either (try it, it's an easy experiment).
 
 

sonarman1
Thanks Noel. I understand the daw is internally handling audio in 32bit or 64bit. ButIhave kept my record, render and import bit depth to 24. So now if I export my mix for mastering in 32 or 64bit format with dithering turned off, will it still benefit? 
If thats possible it will be of great use, I can totally forget abt dithering and make all my exports 32bit.

It will still benefit because irrespective of the source formats of 16/24 bit all internal mixing is done at 32 or 64 bit float. So exporting to a float format is keeping the data at a lossless resolution. With a float format you can literally bounce or process thousands of times with no audible noise build up since floating point errors are infinitesimal.
 
There is no point setting the record depth at anything higher than 24 unless your audio interface has floating point support. Most do not... You should leave the import depth to same as original - that way if you import a float file it will stay in floating point and not get truncated to integer.

Noel Borthwick [Cakewalk]
 
 
It will still benefit because irrespective of the source formats of 16/24 bit all internal mixing is done at 32 or 64 bit float. So exporting to a float format is keeping the data at a lossless resolution. With a float format you can literally bounce or process thousands of times with no audible noise build up since floating point errors are infinitesimal.
 
There is no point setting the record depth at anything higher than 24 unless your audio interface has floating point support. Most do not... You should leave the import depth to same as original - that way if you import a float file it will stay in floating point and not get truncated to integer.

Thanks. Where in the world could I get answers from such knowledgeable people.  Let me consider myself privileged to be here. 
 

 

bvideo

I tried what u suggested. As you said I could only hear only the noise and everything else gets phase canceled. I had to put a limiter with 30db gain to get the noise to audible levels. 

Exactly. Now ask yourself what the audible difference would be between triangular vs. Pow-r 3 dither in a full mix. You could instead insert a satanic message in reverse...no one would know the difference.

Yea, the dither noise was actually much lower in level compared to the noise added by analog emulation plugins. Now I understand why Yoad Nevo doesn't use dither when there is analog noise at the background. 

With 24bit, dither noise is technically impossible to hear till you do digital amplification. If you send it strait into ASIO, it will be way below any modern interface DAC noise level. But if you digitally amplify it, you can shift it into audible domain. Every +6dB shift ~1bit.
 
For myself I have done a noob test. I have checked how many bits (max) should be occupied in a sample of WAV file so I can recognize the presence of dithering on laptop with ~$5 headphones. Lower then 10bit... I am noob and test conditions was bad. So when someone write that is 16bit (CD), I think "well, theoretically possible". But not with 24.