Originally posted by er1kksen Why are the scores so different?

DxO has a score system based on thresholds rather than integration of curves.

E.g., imagine two similiar and very flat curves of SNR vs. ISO, and a threshold where the curves crosses a SNR value, and the corresponding crosspoint ISO value enters the score (actually, this is what they do ...).

Then, the integral over both curves would be very similiar, yet the score would differ a lot.

Mathematically, this is known as an unstable function, where small variations of input can produce large variation of output. Sometimes known as the butterfly effect (when it occurs in iterated functions, not applicable here).

DxO does a decent job with their curves but screws it up with their score function when using a mathematically unstable function.

I've already thought about creating an alternative robust score function and publish an alternative DxOmark figure based on their curves...

Originally posted by PentaxPoke I actually think that the low-light ISO score is relevant and useful.

It isn't wrong but can be very misleading. Cf. above.

I really recommend to use DR over ISO and tonal range over ISO curves as a first guide. If the curves have no flat portion then the low-light ISO score comparison can be useful as well.