Hehe
I dont know that nobody else would be arsed... I once did lengthy tests to analyse the 32 bit mixing engine of nuendo.
It was to prove a point with some twat who simply refused to believe me when I told him you can clip your channel faders to hell and back, and so long as you turn the master down an appropriate amount, nothing will be clipped. He was all like "oh well MR GENIUS pray tell me how the world is flat" - and what made it worse is this guy has ultra-expensive top-of-the-range Apogee A-Ds, which he was keen to boast about, despite apparently not even understanding the principles of digital audio. so I simply had to flatten him with science.
I exported a multitude of projects containing sine waves of various different frequencies - once at unity gain on both channels and master, then several times with channels turned right down and master turned right up, then several times with channels turned up right (and gain plugins inserted!) and master right down.
i'd then normalise, phase invert and mix the results, which always resulted in digital silence - thus proving that in a 32 bit engine it doesnt make a damn bit of difference whether you mix with faders below, around or even way above unity.
at this point things got kind of interesting...
because (without actually admitting to having been wrong in totally discounting my theories) he then altered his argument, claiming that it was a matter of degree. And since I'd only mixed about 4 channels with 12db of gain on each, this wasnt complex enough to notice the errors allegedly provoked by mixing above unity. With a real project of 40 channels, these errors would be significant - he assured me.
So, I investigated a bit harder.
I took my phase-inverted + mixed 'silence' files, and added 40db of gain to each. Now I could see they were not in fact TOTAL silence (altho SF's meters had previously not even flickered, thus making me think they were). They were somewhere between -96 and -105db. And, IIRC, the 'clipping channels' one WAS something 0.3db louder. So in theory, mixing with clipping channels WAS possibly having a negative effect on audio accuracy.
However... I looked at these results and concluded that even if this was the case:
- I never mix with channels constantly clipping by 12db - letting them occasional peaking to 0.3 is more like it
- Even after adding 40db to this "error noise", it was still quieter than the noise floor on the average project studio soundcard.
- Thus, even if the phenomena does exist, its not something i will ever be concerned about
At the time I appealed for users of Fruity, Reason, Logic, et al to run the same tests, because I was curious to see how the different engines stacked up.... nobody ever did tho.