Guys, the DXOMark measurement, I take no doubt, is accurate (up to the equipment precision) for each sensor. But the final scores definitely skewed.
Take an example of ISO performance, the 6D get a score of 2340 while D600 get 2980. But by looking at the SNR18% graph at "Screen", which is how we view our photos in 100% on our monitor, the two sensors produce almost same SNR performance with D600 has a slight leading at low ISO.
But how come the final score for 6D and D600 got almost 1/3 stop difference? There are two things here:
1. DXO use the "Print" to give the final score, which they think this is fair for different sensors which have different pixel count. This is partially correct since when you view a photo from a print out, you won't care about the pixel count of the original sensor, you just care about the noise in the final output. But, there is a "But" here, that is: which re-size algorithm DXO used to get the "Print" measurement? This can be found at their article:
DxOMark - Modeling small pixels
from this article, it clearly states that they use the "average out", quote: "However, the four high-resolution neighboring pixels can be averaged out to form a low-resolution pixel. The statistical formula below shows the noise yield for the downsampled image: ". Yes, in this average out, the noise is greatly reduced, but same applies to the DETAILS. I don't think any downsampling algorithm we used in real life will gain as much noise reduction as theirs.
2. DXO needs the color depth and dynamic range to be a certain level when measuring ISO performance. To be exact, they want the color depth to be over 18 bits when SNR is 30 bits and above. So 6D only got 17.9 bits at ISO 2400 and unfortunately not passing this criteria. so the highest ISO score by this standard is 2340. I don't know why they magically choose 18 bits but not any other value.
So if only consider the SNR value of the two sensor at screen size, and give 30 bits and above as a good ISO performance, both camera will have a score of around 1200.
Lastly, my suggestion is: take a look at the actual measurement in Screen output of each individual segment and ignore the final scoring of the sensor.