DxOMark: EOS 5DS/R sensor is highest-ranked Canon sensor yet
1 Dynamic Range & SNR
DxOMark just published its report on the 50MP sensor in Canon's EOS 5DS and 5DSR. The company's measurements indicate that the 50MP full-frame CMOS is the best-performing Canon sensor to date, offering massive resolution, roughly a 1/3EV improvement in dynamic range over Canon's previous best-performer in this regard, the EOS 6D, with low light performance only a bit behind Canon's low light king, the EOS-1D X. You can compare the 5DS to the 6D and 1D X, Canon's most recent full-frame cameras, here.
The new cameras don't break any records though - according to DxOMark's tests, the EOS 5DSR and 5DS sit in 21st and 22nd place in the company's overall rankings. This is well behind cameras with current-generation Sony CMOS sensors. DxO's results mirror essentially what we found when we put the 5DS R through our Raw dynamic range tests, and you can read our findings here.
The kind folks at DxO have shared their sensor measurement data with us, so we'd like to take this opportunity to comment a bit further on the performance of the 5DS cameras, especially in relation to their nearest competitor, the Nikon D810. But first, a link to DxO's report:
We took the liberty of normalizing 5DS R dynamic range data from DxO to 36MP for a fair comparison against the Nikon D810. This process of normalization effectively ensures we're comparing cameras at a common viewing size, as if you were comparing prints of the same size. This helps cameras with higher pixel counts, which would otherwise be unfairly hurt by their noisier pixels. Since we view pictures, and not pixels, it makes sense to consider noise at the image level, not at the pixel level. By normalizing the 50MP of the 5DS R to 36MP, we're essentially comparing the 5DS R and the D810 on the same, level playing field.
DxO does something similar in their 'Print' mode display of their data; however, that assumes a very low output resolution of only 8MP, which is probably below the resolution most users of these high resolution cameras will desire. Here, we compare at the highest, common denominator: 36MP. Have a look below to see the Raw dynamic range of the cameras, stated in EV, as a function of ISO.
|Raw dynamic range of Canon 5DS R vs. Nikon D810, as a function of ISO. 5DS R dynamic range has been normalized to an output of 36MP, for fair comparison against the Nikon D810. At base ISO, the 5DS R has 11.2 EV of dynamic range, while the D810 has 13.7 EV. That's a 2.5 EV advantage of the D810 at base ISO. By ISO 800, differences are minimal. Raw data courtesy of DxO|
At base ISO, which is 100 for the 5DS R and 64 for the D810, the difference in 'engineering' dynamic range* is a rather hefty 2.5 EV, with the D810 and 5DS R exhibiting 13.7 and 11.2 EV dynamic range, respectively. That means that, down at the darkest levels of exposure on the sensor, tones that receive nearly 6x less light on the D810 will yield similar noise levels to those that received 6x more light on the 5DS R. This explains why you can push and reveal much deeper shadows with D810 Raw files than you can with the Canon files. This ability allows one to expose high contrast scenes for the highlights, which yields a traditional underexposure of midtones and shadows. Due to the low noise of these darker tones recorded on a D810 sensor, one can then push them to make them visible again in a manner one cannot do as effectively on a 5DS R.
Differences in dynamic range between these cameras continue to hold, albeit decrease, all the way up to ISO 800, at which point any differences become minimal at best.
The advantage of ISO 64
There are two things that enable the much higher dynamic range of the D810: first of all, low read noise due to on-chip ADC architecture means your camera's electronics don't have such a high noise floor that dark tones run into them. Second, Nikon implemented ISO 64 on the D810 by actually extending the saturation capacity of pixels (we're not sure how) relative to ISO 100 on previous D800/E cameras. This means that every pixel can capture more light before saturating and clipping. Now if we give the sensors more exposure to take advantage of the increased well-capacities, then darker portions of the scene get cleaner, due to less shot noise and lower risk of running into the, albeit low, noise floor. These are the features that lead to wide dynamic range on the D810, none of which are found in the 5DS cameras.
Having compared the 5DS R vs its nearest competitor, the Nikon D810, with respect to dynamic range, we thought it'd be interesting to see if any of the tones - bright to dark, white to black - of the 5DS R can compete with the best the D810 can do. We can do this by plotting the signal-to-noise ratio (SNR) of every tone the sensor of each camera can record, from raw data provided to us by DxO. All you need to know about SNR is this: the higher it is, the 'cleaner' and less noisy the tone is.
Looking at the SNR plots below, it's clear that the D810 has a higher SNR for virtually every tone the sensor can record compared to the 5DS R. This is due to the technologies we've already discussed that give it high base ISO dynamic range: a higher full-well capacity of pixels means the camera can actually capture more light, while the lower read noise means that at darker tones (on the left end of this graph) suffer less from noise due to the camera's sensor and electronics. In fact, it's the higher read noise of the 5DS R that causes the continuous drop in SNR relative to the D810 for all lower tones (left side of the graph).
|Base ISO signal:noise ratio (SNR) in dB for all tones the 5DS R (blue) and D810 (red) can record. Higher SNR is better. For all tones, the D810 benefits from a higher full-well capacity of pixels, which means the sensor in its entirety can record more light, which reduces shot noise. The lower (downstream) read noise of the D810 also ensures that darker tones have far higher SNR, and therefore less noise, than the 5DS R. This is why the red line pulls further and further away (tones get cleaner and cleaner) from the blue line for darker tones. Raw data courtesy of DxO|
An interesting aside: the high SNR of tones the D810 records, by capturing more light and contributing less read noise, actually makes it a competitor for modern medium format sensors - from Sony itself - found in Phase One backs, and the Pentax 645Z. It's why some of the ISO 64 shots on the D810 just look so sharp and clean and, dare we say, almost have that 'medium format look'. With the D810, Nikon improved on an already class-leading sensor with the introduction of a native ISO 64. But we digress, and will save a SNR comparison between the D810 and 645Z for a separate article.
What's this all mean in the real world? You may remember the tulip sunrise shot from our 5DS R dynamic range analysis. We've included it on the next page of this article. Some of the shadows, including the greens of the tulip stems, in that shot have a Raw signal that translates to a grey level of approximately 0.26 in the graphs above (on an 8-bit, 255 scale). For the 5DS R, that yields a SNR of 6 dB, while the same signal yields a SNR of 18 dB for the D810, which is actually a 16-fold higher SNR (every 3 dB increase is a doubling in absolute SNR). That is impressive, and accounts for the real-world difference photographers see in noisiness of shadows between these cameras.
On the next page, we'll look at Color accuracy and ISO performance
* Engineering dynamic range (EDR) reports the range of tones between clipping and a the lowest tone, on the dark end, that still has an 'acceptable' noise threshold of SNR = 1. Oftentimes this leads to a higher range than is actually usable, as most photographers will not consider tones with SNR = 1 as 'acceptable'. EDR is still useful as a point of comparison between cameras, though quoting absolute EDR numbers may have questionable relevancy. We've been investigating the idea of using a higher SNR threshold on the low end and, indeed, the 'Photographic Dynamic Range' popularized by Bill Claff uses a higher SNR threshold to yield some very useful comparisons and absolute dynamic range numbers, which Bill calculates himself independently on his excellent site.
Oct 1, 2016
Jan 5, 2016
Dec 17, 2015
Aug 4, 2016