High ISO Compared: Sony A7S vs. A7R vs. Canon EOS 5D III
ISO 204,800 and beyond: up to a 2 EV advantage
Sony A7S Raw files show a relatively impressive ability to retain scene detail at *2, especially when restricted to . For cameras such as the A7R and 5D Mark III where these ISOs don't even exist, it is simply not possible to use the images that result from digital exposure boosting in ACR to simulate, say, . Moderate to heavy amounts of noise reduction applied to the A7R and 5D Mark III Raw files at these ISOs does not bring back scene detail that the A7S manages to preserve; rather, this scene detail seems forever lost in noise. But don't just take our word for it - feel free to download the Raw files yourself to see what you can pull out.
What this means is that if you require the extremely short exposures that 6-digit ISOs enable, you'll fare significantly better with a Sony A7S than you will other cameras that don't natively support such high ISOs.
For example, at ISO 409,600 the shadows and darker midtones of the A7S show a nearly 2 EV advantage over the *3 This advantage drops to 1 EV or less in , which are recorded as higher signals at the sensor and are, therefore, less dependent on the electronic performance of the sensor. That is, these brighter regions of the image aren't going to look much better if you decrease the already comparatively low - and therefore not very visible - levels of sensor read noise hanging out in the bright tones. Think of the sensor read noise as hiss on an old audio tape, and recall how this hiss was more salient as a song faded out, but not in the middle of a song where the vocals and instruments were loud enough to drown it out. It wouldn't help much to reduce the hiss in the middle of the song when everything's loud (high signal), but reducing hiss would help as the song faded to quiet (low signal).and - where, at web resolutions (5 MP), the noise levels of the A7S look visually similar to ISO 102,400 on the other cameras.
Similarly, the A7S - with its lower hiss - shows its advantages for very dark tones (low signals), presumably due to lower read/quantization noise or other optimizations that reduce the impacts of read noise (higher conversion gain, for example). Brighter areas, which are only really going to look better if more light is captured, tend to not show much advantage since all the sensors in this comparison collect similar amounts of total light (they're the same size, after all). That doesn't mean that darker tones wouldn't also look better with longer exposures - of course they would! It's just that for brighter tones, collecting more light is the only way to make them look better, and that's why we say they are shot noise *4 limited; that is, their performance is limited by how much total light was captured (and therefore overall sensor size).
Moderately high ISO performance:up to a 1 EV advantage in shadows
How does the A7S compare to the A7R and 5D Mark III at more reasonable - yet still high - ISOs? That 2 EV advantage over the Canon 5D Mark III, at best, at ISO 51,200. Here the darker tones of the A7S look on par with the 5D Mark III's ISO 25,600 when both images are viewed at a normalized 12 MP. The A7R .
The A7S' advantage is - unsurprisingly - nullified for shot noise limited brighter tones, where ISO 51,200 yieldsacross cameras. This leads us to our next point...
Lower ISO and brighter tones: it's a wash
Before we get too carried away by the seeminglyof the A7S over the A7R in high ISO shadows, let's take a look a different region of the image where . What's interesting about this particular crop is the gradient - from dark to bright tones - as you move from the top to the bottom of the crop. And you may have been catching the drift - from the previous two sections - that noise advantages can vary as you go from dark to bright tones. So how do the A7S and A7R across various tones at ISO 51,200?
First, you'll note that while the A7R's shadows (top of the crop) are worse than the A7S', brighter tones at the bottom of the crop (the street surface lit orange by the street lamps) are roughly similar in noise performance. This results from the simple fact that brighter tones are recorded as higher signals on the sensor and are, therefore, shot noise limited. That is, their noise performance is determined largely by sensor size and less so by sensor read noise, assuming focal plane exposures are matched and barring large differences in sensor efficiencies. Shadow noise performance, however, is influenced by both shot noise and sensor read noise, and it's(scroll back up to widget) that we see the detrimental effects of a sensor not tailored for sensitivity as well as the A7S is. Though we're admittedly speculating at this point, we'd imagine the A7R's drop in performance relative to the A7S for these very low signals to due to either (1) the higher cumulative read noise resulting from the three times as many pixels that need to be read off the A7R compared to the A7S, and/or (2) conversion gain or other sensor-level optimizations on the A7S aimed at more precisely representing very low signals.
Further on(scroll back up to widget), performance across cameras pretty much evens out. In the absence of shadow boosting, this is to be expected, as noise performance for well-exposed images is largely determined by sensor size (for cameras of similar generations).
If we were to sum up our thoughts thus far on the (Raw) high ISO performance of the Sony A7S, it would go something like this:
The Sony A7S enables the use of incredibly high ISOs, including some unavailable to most cameras. When shooting at incredibly low light levels, the A7S will most likely give you significantly better results than any camera that does not natively shoot at these ISOs. There are also some advantages to be had over the higher native ISOs of the cameras we tested: we see the A7S overtake the A7R in normalized shadow noise performance at ISOs above 6400 in our tests. At more moderately high ISOs (6400 and below) as well as in brighter regions of images - where image quality is determined primarily by how much total light is captured - the normalized ISO performance of the A7S will be similar to that of full-frame cameras of its generation.*5
Do the results surprise you? Or do they simply corroborate what you expected? Feel free to discuss and let us know your thoughts in the comments below.
*2 Please recall that since the A7R and 5D Mark III don't offer ISO sensitivity settings above 25,600 and 102,400, respectively, these higher ISO shots were simulated by maxing out the ISO on each respective camera, adjusting shutter speed, and then digitally boosting exposure in ACR to aid comparisons against the higher ISOs of compared cameras. For example, ISO 51,200 on the A7R was simulated by leaving the camera at ISO 25,600, but giving it 1 EV less exposure compared to the ISO 25,600 exposure. ISO 102,400 was simulated by leaving the camera at ISO 25,600 and giving it 2 EV less exposure. And so on and so forth. Therefore, you'll have to brighten these files accordingly yourself - else, they will appear underexposed.
*3 Note that the A7R was digitally boosted 2 EV, which is likely the reason for the brighter shadows. Noise levels, however, appear comparable between this digitally-boosted ISO 102,400 on the A7R and ISO 409,600 on the A7S. Certainly more comparable than , anyway.
*4 Shot noise is a type of noise whose contribution decreases the more light that is captured. It's one of the main reasons lower ISO images - with their longer exposures - look less noisy and cleaner than higher ISO images. You can read more about shot noise (also known as photon noise) here.
*5 As always, there are caveats. Although we say that noise performance for 'moderately high' ISOs are primarily determined by how much total light is captured (and, therefore, the relative shot noise), a certain type of sensor noise known as upstream read noise can still typically have an appreciable effect. This is because this source of noise - typically constant across reasonable exposures - is amplified more and more at higher ISOs. Therefore, lowering this particular source of noise can help noise performance at most, especially higher, ISOs. However, when we say it's the total amount of light captured that predominantly determines ISO performance at moderately high ISOs, we're assuming relatively constant upstream read noise performance for sensors of the same generation.