What is and what isn't 32bit float?

[OldTimey]

david - everything you said here is perfectly obvious to me, and your 'plastic bits in water' analogy is plain silly.

i'm talking about processing done to signals before it gets written to disk. An 828mk3 can take a 24-bit sample, apply compression to it in the 32-bit sphere (again DP is nowhere in this process), and then send that 32-bit piece of data to DP to be recorded as either a 16-bit (truncated), 24-bit (truncated), or full resolution 32-bit float (1 bit for sign, 23 bit mnantissa, 8-bit exponent) sample. In the latter scenario, this string of ones and zeros is recorded to disk bit for bit when working at 32-bit. Are 828mk3's actually capable of this? I don't know. But it would seem rather easy to pull off. When played back through DP's mix engine, yea it all gets upsampled again if needed.

to me, the biggest and really only upside of tracking to 32-bit float is not having to truncate any onboard processing your interface might provide, if that interface is even capable of sending a 32-bit sample to your DAW.

[David Polich]

david - everything you said here is perfectly obvious to me, and your 'plastic bits in water' analogy is plain silly.

i'm talking about processing done to signals before it gets written to disk. An 828mk3 can take a 24-bit sample, apply compression to it in the 32-bit sphere (again DP is nowhere in this process), and then send that 32-bit piece of data to DP to be recorded as either a 16-bit (truncated), 24-bit (truncated), or full resolution 32-bit float (1 bit for sign, 23 bit mnantissa, 8-bit exponent) sample. In the latter scenario, this string of ones and zeros is recorded to disk bit for bit when working at 32-bit. Are 828mk3's actually capable of this? I don't know. But it would seem rather easy to pull off. When played back through DP's mix engine, yea it all gets upsampled again if needed.

to me, the biggest and really only upside of tracking to 32-bit float is not having to truncate any onboard processing your interface might provide, if that interface is even capable of sending a 32-bit sample to your DAW.

Well if you want to continue discussing lofty math and processing
concepts, that's cool with me. I'm not enough of an authority to dispute
whatever it is your claiming. Believe what you want.

All I know is, for me, this is a completely moot discussion anymore. 24-bit is fine for me, I have no need or desire to bother with keeping
things at 32-bit, and frankly I don't think bit-depth has anything to do whatsoever with whether a track has any impact on people.. A good song, good performances, a good mix..that counts light years beyond what your bit depth is. Mostly it is the song that counts above anything else.

I'm not saying this with any condescension, or combative attitude or disrespect. I just think that at the end of the day, it is an exercise in
theory application and really doesn't have anything to do with whether you
can mix a record well.

[Mark Walmsley]

I can add nothing scientific to the discussion about the math of 32 bit files BUT when DP first added the 32 bit file option (I think it was at v6?) I did some listening tests which were interesting.
I had always felt that a 24 bit looped back mix (through FF800) sounded a little less 3 dimensional than the fantastic internal DP sound. When I listened to the new 32 bit format mixes I could no longer hear a difference. This was particularly noticeable to my ears on sparse ballads with say 1 vox an acoustic guitar and a reverb tail.
I have stuck with 32 bit ever since and then use Izotope RX to change bit rate when doing my own quick and dirty masters. Don Bartley at Benchmark mastering has no problem receiving 32 bit mixes.
I met Mike Stavrou a while back and asked him whether I was imagining this and he thought it possible that extra detail in quiet mixes could be captured this way.
Please take it all with a grain of salt, but it works for me!

[OldTimey]

i dont use 32-bit float, i do half of my tracking on cassette tape--but i am a man of science and like to know how stuff works--i think the OP wanted to know how 32-bit float is useful and how it works, and I came up with a scenario where, mathematically speaking, recording at 32-bit will possibly store different data than when recording at 24-bit. That some don't care about this is of really no consequence to the fact that it may happen, and of really no interest to me as it pertains to this thread. I've seen enough "it's the song/performance/ears-not the gear" kinds of replies to threads that another one is just noise to me...