AUP and AUP3 project files are more complex and more “fragile” than WAV files.
You should back-up as WAV or FLAC immediately after recording. I doesn’t need to be 32-bits to preserve all of the information. It just has to match (or be higher than) the bit-depth of your interface (16 or 24-bits).
Depending on what you’re doing, you may not even need a project file. But it can serve as a backup if that’s the only backup you’ve got.
Audacity uses 32-bit floating-point internally. It makes processing “easier” and floating-point can go over 0dB so if you do anything that pushes the peaks over 0dB (like boosting the bass, for example) you won’t get clipping (distortion) as long as you lower the level before exporting.
And/or you can export as 32-bit floating point WAV but your “final production” shouldn’t go over 0dB because it can clip the listener’s DAC.
An AUP project contains multiple files and the actual audio isn’t in the AUP file.
There are 8-bits in a byte so one 32-bit sample is 4 bytes (8 for stereo).
(32-bits/8) x 48,000 x 2 channels = 384kB per second. (That doesn’t include any project overhead, or embedded artwork in a WAV file, etc.) That’s a bitrate of 3072kbps. (Usually we only “talk about” bitrate for compressed audio.)
1.2GB should be somewhere around 2-hours.
The “pro studio standard” is 24-bits/96Khz. But CD quality is generally better than human hearing.
A proper scientific, blind, ABX Test can be “humbling”. IF you can reliably hear a difference, you’ll have to listen VERY carefully.
…A good quality MP3 (lossy) can often sound identical to the uncompressed original (in a blind ABX test), or again, you’ll have to listen carefully to certain parts of the recording to hear the difference. (A low-quality, low-bitrate, MP3 can sound pretty bad!)
The exception could be a floating-point file that goes over 0dB and then the integer copy would clip.