Exporting a mix: 24-bit vs 32-bit float

When you’re exporting a mix (aka "bouncing"), you have some potentially confusing options to muddle through. Here’s one that can trip people up: should you export your mix at 24-bit, or at 32-bit floating point?

Most people know that 24-bit has been a standard for the last 20 years or so. But, 32 bits seems ... better, right? It's 8 more bits, after all!

The answer: you can (and should!) do your exports at 24-bit. 

The reason might surprise you: as long as the peak level isn’t going over 0 dBFS, which it shouldn’t be because you should always be working with a limiter on your master bus, there’s no sonic difference between 24-bit and 32-bit float!

But how can that be?

It has to do with how floating-point data works. And here is something that might surprise you: there are 24 bits of audio data in both 24-bit and 32-bit float!

But ... what about those extra 8 bits in the 32-bit floating-point data? So, here’s the thing about them: those extra 8 bits are not audio data — they’re an *exponent*.

This exponent allows for multiplying the audio data out into way bigger dynamic ranges, with amplitudes far above 0 dBFS — which you are not using, and indeed can not use, because you are using a limiter on your master bus. 

So when you export a mix as 32-bit floating point data, it's literally just 24-bit audio data with 8 zeroes tacked on the beginning! And what's the point of that¹.

Float on — jamie

1. To be perfectly clear, there is no harm in exporting your mixes as 32-bit float. Those extra zeroes aren’t going to, like, ruin your mix. But we’re doing engineering here … and it’s good for us engineers to know technical things! And now you know.