Analogue amplification (pre-digitization) is the best way to implement a variable ISO sensitivity. When you do so you are only amplifying the photon noise. When you amplify the digital data (i.e. by multiplying all the values by a fixed amount) you amplify the photon noise, the read noise and the quantization noise, leading to more noise overall!
Most Canon DSLRs only have analogue amplification circuits for the whole stops (100, 200, 400, 800 etc.), when you select one of the intermediate fractional values (520, 640 etc.) uses the closest analogue amplification stop and then uses digital multiplication to give the correct overall sensitivity. E.g. ISO640 is really ISO800 multiplied by 0.8, ISO500 is really ISO400 multiplied by 1.25 etc.
This is bad as you either end up using a higher analogue ISO and lose highlight headroom, or a lower analogue ISO and get more noise than necessary. For this reason I would advise against using the fractional stops on Cameras that support it.
For an example of how bad digital amplification can be compared to analogue, see this image:
[![][1]][1]
The top image was shot at f/2.8 1/30s and at ISO1600, i.e. 5 stops of analogue amplification. The bottom image was shot at an identical f/2.8 1/30s but ISO100, and then 5 stops of digital amplification were applied in Photoshop. The result is that digital amplification gives a lot more noise. [1]: https://web.archive.org/web/20150913050221/http://www.mattgrum.com/photo_se/analogue_vs_digital_amplification.jpg
