23 February 2026 | 6 min read
24 bit vs 32 bit float: what actually changes on set
The marketing says you can stop riding gain. The physics says something slightly more nuanced.
32 bit float recording arrived with a bold promise: never clip again, never set a level again. Like most bold promises it is mostly true, with an asterisk that matters.
The difference in one paragraph
24 bit fixed point stores samples as integers with a hard ceiling. Exceed it and the waveform is clipped, permanently. 32 bit float stores samples with a mantissa and an exponent, giving a dynamic range so wide that in practice nothing in the audible world overflows it. A signal that looks clipped can be pulled back down intact.
Where the asterisk lives
The float format protects the recording, not the microphone or the preamp. If a loud source overloads the analogue input stage before conversion, the distortion is baked in and no amount of floating point range recovers it. Good 32 bit float recorders use dual analogue to digital converters at different gains to widen that window considerably, which is why the promise mostly holds.
- Sudden gunshots, doors and shouts: 32 bit float saves the take.
- A lav buried under clothing with a screaming subject: still risky at the preamp.
- Very quiet ambience: float means you can lift it later without touching the noise floor at capture.
What it changes in your day
It does not remove the need for sensible gain staging, but it removes panic. You set a reasonable level, you stop chasing peaks during a take, and you fix the outliers in post. On a run and gun documentary that is a real change in how the job feels.
What to deliver to post
Here is where crews get caught out. Many post pipelines still expect 24 bit fixed files. A 32 bit float file dropped into an unprepared session can import at the wrong level or be rejected outright.
The safe habit is to archive the 32 bit float masters and deliver 24 bit versions with the levels already normalised sensibly, unless post has explicitly asked for float. Ask before the shoot, not on delivery day.
Converting cleanly
Converting float to 24 bit is a real decision, because you are choosing where the ceiling lands. Set the level first, then convert. Every export on this site lets you pick the output bit depth explicitly, so you decide rather than discovering later what a default did.