I've never been quite comfortable with the results from the A7 I have now pushed any more than 4 stops. Then again, I never push my 5D3 shadows more than a stop without using MagicLantern.
Interesting. Actually, thanks for this data point. I was mildly considering getting rid of my A7R and getting the A7 instead, b/c of shutter shock on the A7R which makes the 70-200 FE of limited utility, and b/c phase detection on the A7 actually helps nail focus faster, and even track to a certain degree. AF-C shouldn't even be offered on the A7R, in my opinion, but that goes without saying for most CDAF-only systems.
But now your concern added with the slightly worse ISO performance according to both DPR studio scene and DxO makes me think I'll ditch the A7 system altogether until Sony introduces a model with electronic first curtain in the 36MP sensor (and hopefully stops compressing their RAWs).
The spikey behavior in the histogram represents posterization. This isn't even really a function of the sensor at all, just that we don't have enough information to provide continuous color contours. It's not going to appear as clearly as your exampe though due to the random noise. In any case, Sarangiman clarified that he meant that higher ISOs were emulated with higher base ISO rather than always sticking with ISO 100 which, surprisingly to me, actually makes a huge difference.
Yeah, but then you could call the A7S 'posterized' at ISO 409k, where it's trying to make an image with ~50 photons or less per pixel (50 photons = white), according to sensorgen's back-calculations. At least it's dithered. But yes what you're talking/worried about is a form of quantization error, since at ISO 100 you're not counting every electron. This'd be largely fixed with a 16-bit ADC, though. So hopefully we'll see those soon in Sony/Nikon cameras.
And I think by ISO 400 you're counting every electron, which is why you see a difference. You're also amplifying even the lowest signal 4x, so that may be just enough to make it less affected by the 2-3 electrons (my guesstimate) of downstream read noise that even Exmor sensors have.
The whole point was that even Exmor RAW files can't be pushed 6 stops without trouble.
Er, I'm going to have to beg to differ. Here's the D810 at ISO 6400 and at ISO 100 pushed 6 EV:
Can you tell which is which?
Btw, in the ISO 100 file, that road has a signal of average 7
for the green channel, where the SNR is 3.5
. Red channels is like 4
. These are all on a 16-bit scale. So pixels with signals of literally 4-7 in the Raw file (1 being minimum, ~16,000 being maximum).
So I'm not even talking about pushing a midtone or even a shadow +6 EV. I'm talking about pushing some of the deepest of the deep shadows 6 EV. With brighter tones - you can't tell *any* difference between ISO 6400 and ISO 100 pushed 6 stops. But here even with tones all the way down at the floor of the sensor/Raw file, it's hard to see the difference. If that doesn't wow you...
Point being: yes you can push 6 stops for certain Exmor sensors. Though it does get tricky with ACR, since it's not really built to do that. I had to use the 'Blacks' slider which ends up reducing contrast, and I adjusted until I got the same brightness in the road as the ISO 6400 file.