I ran the same series of tests that I’ve run on the D4, D800E, and M9 on the Sony RX-1.
I expected to find more signs of self heating of the sensor, since the RX-1 runs with live view on all the time, but I did not. Here’s a series of closely-spaced exposures of 1/4 second at ISO 6400. The subject is total darkness. The dotted green lines are least-squares fits to the green data.
And here’s the standard deviation of the series:
The noise floor versus ISO test did produce a surprise:
The values for the ISO 100 setting don’t make sense. However, they are repeatable. I had lens chromatic aberration and falloff correction turned on in the camera. Maybe that has something to do with it. I may check to satisfy my curiosity, but I can’t see any way this anomaly could affect picture making.
The signal to noise ratio (SNR) plotted as a function of ISO setting at an exposure calculated to give a constant 4000-count average green histogram measured in a 200×200 pixel central sample drops off as (by now) I would expect.
Subtracting out the half a stop per stop falloff from reduced photon count as the ISO goes higher, we get this (the narrow lines indicate plus and minus two standard deviations as calculated from the data from the 16 exposures per data point):
There are very small mid and high value SNR improvements to be had in increasing the ISO settings of the camera over pushing in Lightroom or ACR. After 800, you might as well push in ACR or LR. Over 1600, you’re actually hurting yourself fractionally by turning up the ISO setting.
In the shadows (green counts of around 500), the SNR plot looks like this:
And, with the half-stop/stop slope subtracted out, it looks like this:
In the shadows, you can’t help yourself much by increasing the ISO setting, but, past 1600, you can do yourself a bit of damage, especially in the red channel.
[…] stop getting any advantage in turning up the ISO on the RX1 at ISO 800 or ISO 1600. ETTR — Just crank up the ISO? Part 14 | The Last Word There's some digital gain going on that means that you want to avoid ISO 6400 and above. […]