Is 32-bit float the best "project" setting?

16 and 24-bit Wave files store the audio data as signed integers.

The Cubase audio engine (as most DAWs these days) processes the audio as 32/64 bit normalized floating point.

Converting from integer to normalized floating point requires math computation; it takes CPU power, has round off error, etc.

So, at what point in the process are all the wave files converted from integer to floating point? Are the files converted when the project is opened - and all the converted files stored in memory (using RAM/HD) - or are the files converted on the fly as they are being read (using CPU resources)?

Bottom line question: would saving all files, and setting the “project” as 32 float, avoid all this converting and reduce CPU load?

The files are written to disk at the project bit depth. Any files that are not 32fp are converted on the fly at the point of first processing (gain change, fx, etc). Any disk writing operations are written at the project bit depth.

Converting bit depth are one of the things that processors are very good at. No real gains to be had in processor performance. Having said that I usually just set the project bit depth to 32fp and leave it there as Disk overheads are so vast these days that again you’ll never notice. SSD space requirements excepted!