sploo wrote in post #17363546
Your first paragraph is pretty damning, but I probably wouldn't disagree. Though the dual pixel AF is interesting to some shooters.
For the high iso noise concerns; remember that noise per unit area is what matters, not noise per pixel. Thus it would still be a step forward to have a high MP sensor that gave more detail for low iso/good light and the same noise per unit area as a lower MP predecessor at high iso.
I've not looked into sensor stats for the 7D2, but does it scale in abilities as expected vs a full frame sensor? I.e. if a FF sensor were made from it, would it match the 5D3 for high iso noise? If so, it would be better than Canon's current offerings for some shooting styles.
If you go the DxOMark graph route, all you need to do is slide the 7D2 results for SNR18 up by 4dB to see how it would perform filling a FF with its pixels, and up 2dB to see how it would do as a 1.3x crop. For DR, you'd slide the 7D2 up 2/3 stop as FF, and 1/3 stop as a 1.3x crop. These three sensor sizes are pretty close to having equal spacing in noise characteristics, spread over 2/3 stop, with AOTBE. IOW, filled with the same pixels, a FF camera has 1/3 stop less image-level noise than an APS-H, and 2/3 stop less than APS-C. And of course, if filled with the same pixels and quality of readout electronics, the noise per unit of sensor area, or focal-length-limited noise is exactly the same.
What you get if you slide the trends like that to create something like a "Surface" (noise quality per unit of sensor area) option derived from DxO's "Print" option, is that the 7D2 has superior SNR18 over all FF and APS-H cameras by at least a small amount (the 1D3 and especially the 1D2 and original 5D fall way behind), surpasses all larger sensors at base ISO DR, and is about in the middle of the 5D3/6D/1Dx spread at high ISO DR. Of course, DxOMark is not without its faults. It measures noise as a monolithic statistic, with no regard for noise character, but this actually handicaps the 7D2 compared to many other Canons in the graphs, because the 7D2 has great noise character, and the poorer character of some other cameras makes their performance a little worse than the graphs suggest (the 5D2 is an excellent example - it's noise quantities at high ISO were very good in its day, but its noise character at both high and low ISOs was horrible - the 5D3 is a lot better, but not as good as the 7D2 or 6D in character).
Of course, just as the 7D2 has less high-ISO read noise than the 70D with the same size pixels, the same size pixels on a future camera could have even less pixel-level high-ISO noise, and equal or better the 1Dx in a FF. There is still a long way to go in reducing read noise; the target is not 1 electron, it is 0.15 electrons, because if the read noise is less than that, it can be removed from a capture without removing any photons captured, because each photon quantity has its own distinct bell curve in a histogram, not overlapping with the others. It is no easier to get read noise down to 0.15 electrons with large pixels; in fact, cell phone camera photosites have the least read noise as measured in electrons, and are closer to eliminating read noise completely.