According to 3.4.2. section "data precision" on [this website][1], the Canon 450D has 14bit deph. Canon CR2 stores RAW data as JPEG Lossless format.

 - Image size: 4312x2876 
 - CFA: 4313x2877 (cambridge in colour sensor
   scheme)

As 14bit depth `(2^14, 16,384 variations)` are stored as unsigned __int16 `(max value 65,535)`
"unsigned __int16" has size of 2 bytes.
24,802,624 bytes =~ **24mb per file** (IFD#3 only)

We could have data precision in 64 bits:

As 64bit depth `(2^64, 18,446,744,073,709,551,615 variations)` are stored as 
unsigned __int64 `(max value 18,446,744,073,709,551,615)`
"unsigned __int64" has size of 8 bytes.
99,210,496 bytes =~ **100mb per file** (IFD#3 only)

I'm not saying JPEG Lossless format stores information in unsigned __int,
but I found out the normal JPEG uses a widthxheightx3 matrix in YCbCr color space that can be converted using libjpg.

Not talking about "you couldn't tell the difference", the point is about 
we being able to capture some sort of a ultra high dynamic range image which only blown up thing would be pointing to the sun directly or bulb mode exposures.

Maybe I am misunderstanding something, who knows. The Canon 450D seems to have a **128MB buffer** by the way.


  [1]: http://lclevy.free.fr/cr2/