>It is remarkable if Nikon just manages to achieve the D7000 >ISO performance at 24 MPs.
They've already done that with the D5200.
Making a sensor higher-MP doesn't affect noise performance that much. What you lose in noise performance of the individual pixels you make up in the fact that you are averaging the noise out over more pixels.
What matters is the amount of noise in the output image, not the amount in the individual pixels. That's determined by how much noise is generated in the sensor and how much light it gathers. Overall noise generation is determined by the sensor technology; how much light is gathered is determined by the size of the sensor. Megapixels aren't really a significant factor.
>Personally I think the well is >likely tapped for ISO performance on DX by and large. The >reality is that the sensor size is likely at the point of >bottlenecking substantial upgrades.
I think so, too. Barring any breakthrough technologies, that is. (Maybe this one?)
>Some have suggested this methodology is flawed as the numbers >are fairly arbitrary. Note what is does not take into >account, how higher ISO photos look. What happens when you >try and recover detail from the shadows, etc.
DXOMark's approach is arbitrary, but I find that it pretty well matches my arbitrary standards. There is no real line-in-the-sand measurement for how much noise is acceptable, but having a consistent basis of comparison is useful. If you find that the DXOMark numbers are consistently low or consistently high for you, adjust accordingly. I would be surprised if you find them very inconsistent, however. At the level of fine detail, there may be differences between sensors and processing systems caused by such issues as color noise vs luminance noise that impact shadow detail, but for the most part those are second-order effects to me.
>I personally don't find that either the D3200 or D5200 have in >actual fact, better ISO performance than the D7000.
Which matches up nicely with the DXOmark numbers -- that 1/7-stop difference I mentioned as being essentially indistinguishable. As you go to higher ISOs where the SNR drops, the performance of the systems may diverge a bit, but probably not that much.