I do mostly post, and I'm about to record sound on a friend's short with a recorder that does 32-bit float. People tell me gain doesn't matter anymore. That sounds too good. What does 32-bit float actually save, and where can it still go wrong?
It saves the file. A 32-bit float recorder (with dual converters, like Sound Devices' design) can capture peaks above 0 dBFS, and you can pull them back down in the DAW without distortion. Quiet passages can come up without the noise penalty of a low 24-bit recording. It does not save anything before the converter. A mic capsule driven past its maximum SPL distorts. A wireless transmitter that only works in 24-bit can clip, because 32-bit float has to exist through the whole chain. Wind, a bad angle and a noisy room are all still recorded faithfully. Files are also about a third larger than 24-bit. So set a sensible level, mainly so your headphone feed and the mix track make sense. Source: Sound Devices, 32-bit float files explained, https://www.sounddevices.com/32-bit-float-files-explained/
Editor's angle: ask your post house before the shoot whether they want 32-bit float or 24-bit files for editorial. Some pipelines handle float fine, others want a 24-bit copy for the cut and the float originals kept for the sound team. Better to know on day one than to transcode a hundred files the night before turnover. Put the answer in the sound report header so everyone downstream sees it.
Thanks. Recorded float, set levels as if it were 24-bit, and only needed the extra headroom once, when an actor slammed a door nobody had rehearsed. That one take alone justified the recorder. The rest of the shoot it behaved like any good 24-bit machine, which is the point: the mic placement and the quiet room did the real work. Delivered 24-bit copies to editorial as Ilya suggested.