32-bit float vs 24-bit for Recording – Hosts & Applications (Sequencers, DAWs, Audio Editors, etc.) Forum

* 24bit int is sometimes stored in 32bit words, because that is more efficient in terms of speed. Shoving four 3-byte samples into three4-byte words is an extra computation. You need to test it per DAW, do some simple calculations to check it.
* A 32bit floating point number is technically a 24bit integer with a 8bit exponent. To increase the range above the upper limit and below near-zero.
* In the range used for audio recordings, a 20bit integer is enough. No A/D or D/A converter goes beyond that dynamical range.
* imho even 16bit suffices if you have proper control over your levels aka The Art of Gain Staging.