A:M implements the most accurate compression schemes of OpenEXR, which i believe use 32-bit floating point numbers for each channel (except PXR24)
A 32-bit floating point channel should make for a very, very long gray scale.
A:M doesn't implement the "lossy" B44 and B44A, AFAIK.
You can read about all OpenEXR's compression options in this PDF under "Data compression"
I dont know of any format that uses a 48-bit integer for each channel, is there one?