Nikon D850 Review

Image Sensor

As I have already mentioned on the first page of this review, the D850 features Nikon’s first ever back-illuminated (BSI) CMOS sensor, which is kind of a big deal, since BSI sensors are more efficient than traditional CMOS sensors in capturing light. Since photodiodes are positioned closer to the sensor surface on a BSI sensor, it minimizes light losses, which results in better image quality, as shown below:

So despite having a higher resolution sensor, the Nikon D850 should be able to deliver outstanding image quality that surpasses the image quality of its predecessor, something we will be able to compare and see in the next pages of this review. It is also worth mentioning that Nikon designed the D850 sensor just like it has done before on many other DSLRs and used a third party silicon foundry to manufacture the chip. Many thought that the D810 replacement would again have a Sony-designed sensor since both the D800 / D800E and the D810 / D810A had Sony sensors, but Nikon decided to take a different route this time, which in my opinion is good news since we probably won’t be seeing other cameras with the same sensor any time soon (the original Sony-designed 36 MP sensor was used on the Sony A7R, Nikon D800/D800E/D810/D810A and on the Pentax K-1). Plus, Sony already hinted that it will be keeping the best of its sensors for its own cameras, so it was about time for Nikon to move its focus towards other sensor manufacturers.

Image Quality and ISO Performance

Due to the structural and technological improvements to sensor design, Nikon was able to push the sensor performance of the Nikon D850 to its limits. Nikon has kept the amazing dynamic range performance of ISO 64 and although I wish the native ISO sensitivity was lowered even further, unfortunately, current sensor technology simply does not allow yielding even more dynamic range out of modern sensors. Based on what I have seen so far, dynamic range performance differences between the D850 and the D810 are minimal, but noise performance differences at high ISOs are definitely noticeable – something I will be presenting in the next pages of this review.

Overall, it looks like the Nikon D850 shows yet again class-leading performance in sensor performance, which is exciting to see. However, my biggest surprise with the D850 so far is not in the image quality department, but rather in the implementation of mRAW and sRAW, which I will cover below.

mRAW / sRAW Formats

As you may already know, the Nikon D810 already had an sRAW format available as an option in the camera menu. The problem with that implementation, as detailed in my sRAW Format Explained article, was that Nikon was stripping out a lot of data from the sRAW file and color + luminance information was being demosaiced by camera software to form RGB pixel data. This resulted in very flat images that barely measured 11 bits of actual data preserved in those images, making the sRAW format somewhat useless.

When the Nikon D850 was released, Nikon hinted at the camera being “three in one”, referring to the ability to choose between three different resolutions (full RAW, medium RAW / mRAW and small RAW / sRAW) for those who needed the ultimate flexibility. However, after seeing the sRAW implementation of the D810, I was very skeptical of this statement and thought that we would get yet another bad small RAW implementation. After exchanging a couple of emails with Iliah Borg, the genius behind the amazing FastRawViewer (see our detailed review) and RawDigger, who told me that the mRAW and sRAW files were implemented differently on the D850, I decided to do my own tests to see where mRAW and sRAW stands on the D850.

First, let’s talk about some technical information that Iliah revealed to me in regards to the wrongly stated resolution numbers by Nikon. Below are the Nikon specifications for each RAW format size:

• Full Size RAW (RAW Large): 8,256 x 5,504 pixels (45.4 MP, 100% resolution)
• Medium RAW (mRAW): 6,192 x 4,128 pixels (25.5 MP, 75% linear resolution, 56.25% total resolution)
• Small RAW (sRAW): 4,128 x 2,752 pixels (11.3 MP, 50% linear resolution, 25% total resolution)

It turns out that the reality was actually quite a bit different. Take a look at the below resolution numbers:

• Full Size RAW (RAW Large): 8,288 x 5,520 pixels (45.7 MP. 100% resolution)
• Medium RAW (mRAW): 7,104 x 4,728 pixels (33.6 MP, 85.7% linear resolution, 73.4% total resolution)
• Small RAW (sRAW): 6,216 x 4,136 pixels (25.7 MP, 75% linear resolution, 56.2% total resolution)

It turns out that Nikon performs double resampling to get to mRAW and sRAW sizes. First, the camera performs RAW data resampling to get to 7,104 x 4,728 and 6,216 x 4,136 resolution from the full size 8,288 x 5,520 RAW file. From there, the images are again resampled by RAW converter software to get 6,192 x 4,128 (mRAW) from 7,104 x 4,728, and 4,128 x 2,752 (sRAW) from 6,216 x 4,136. This is evidenced by looking at the actual dimensions for both mRAW and sRAW files using RawDigger:

This means that Nikon is actually storing 33.6 MP worth of information on an mRAW file, rather than the claimed 25.5 MP, while the sRAW file actually contains a total of 25.7 MP resolution. I tested the validity of this claim by Iliah Borg by extracting a TIFF file from RawDigger and indeed, the above-mentioned numbers came out to be true. Just because Nikon wants to keep the resolution of the mRAW and sRAW files small, it instructs RAW converter software to perform the second phase of resampling, which then reduces the files to 25.5 MP for mRAW and 11.3 MP for sRAW.

The above is further supported by the actual size of RAW images. Let’s take a look:

• Full Size RAW (RAW Large): 58.2 MB (100%)
• Medium RAW (mRAW): 32.7 MB (56.2%)
• Small RAW (sRAW): 24.8 MB (42.6%)

A couple of interesting points here. The above file size data for the Full Size RAW is with Lossless Compression option enabled. shooting in 14-bit. With Uncompressed RAW, the file ended up at around 100 MB in size. However, no matter if I picked Lossless Compression or Uncompressed RAW, the mRAW and sRAW files always produced the same size files, which shows that Nikon is always applying compression to those files – no surprises there, since we already know that from the D810. Now if you are wondering what type of bit rate Nikon is pushing on those compressed RAW images, then you should know that it is always 12-bit.

However, pay close attention to the file sizes relative to the full-size RAW – they are still too big. While the mRAW file size closely matches the change in resolution, it is purely coincidental in this situation – if it wasn’t for the extra 12-bit compression, it would have been roughly 73.4% of total resolution. But the sRAW file is what gives it away. Technically speaking, if Nikon implemented proper pixel binning, the sRAW file should have been around 25% of the total file size. However, we are looking at 42.6% file size, because the actual sRAW contains more pixels than it should and 12-bit compression cannot bring the file size beyond that.

You might be puzzled as to why Nikon decided to push double resampling instead of just keeping the camera-sized 33.6 MP and 25.7 MP files. My guess is that the files didn’t look good, since Nikon is not performing true hardware-level pixel binning, but rather some kind of pixel skipping. Take a look at the 100% crops of full-size RAW, mRAW and sRAW as stored in the file (converted to TIFF via RawDigger and then cropped in Photoshop):

Yikes, those mRAW and sRAW files lack a lot of detail when compared to the full-size RAW file! Plus, if Nikon did not perform the second level of resampling, mRAW and sRAW files would contain the same amount of noise as the full-size RAW file when shooting high ISO, bringing the benefit of mRAW and sRAW even further down.

Another important point to keep in mind is that there is a pretty big overhead when it comes to creating mRAW and sRAW files. This is evidenced by the Buffer Capacity comparison between the three formats:

That’s obviously not good – going with mRAW cuts the buffer in half, whereas sRAW brings it down even further to almost a quarter. That’s a pretty big penalty for sure! Now those numbers apply only when using the fastest XQD cards because when I tested with a slower UHS-II SD memory card, the numbers were very different. With full RAW shooting in 12-bit lossless compressed format and at ISO 100, I got a total of 51 images before the buffer filled up. Changing to mRAW resulted in 67 images, whereas sRAW yielded a total of 65 images. So unless you shoot with a very fast XQD card, you can expect mRAW and sRAW to increase the buffer size by a tad, but not by a whole lot. This shows the importance of using fast media – if you want the D850 to perform, you must use the fastest XQD cards out there.

Despite all this, one big question remains – why is Nikon doing all this resampling on the D850 and why didn’t it use the same process as sRAW on the D810? The next section is going to reveal that, which for some photographers is going to be very exciting!

mRAW / sRAW Format Surprises: Shadow and Highlight Recovery

The next task was to determine what type of damage the new mRAW and sRAW format would exhibit on real-life images. Based on the prior experience with the Nikon D810, we know that the sRAW format on that camera was horrendous, creating images that had less than 12 bits of data – it was easy to see how damaging sRAW was when recovering shadow and highlight detail. I decided to perform a similar test on the Nikon D850 to see how mRAW and sRAW dealt with data and below are my findings. First, let’s take a look at a scene captured by the Nikon D810. We will first recover shadow details by 4 full stops (click to enlarge):

It is very clear that the Nikon D810’s sRAW format is very poor at storing even 12-bits of data – you can clearly see that the sRAW format cannot recover the full 4 stops like the full-size RAW file can. Now let’s take a look at what happens when an image is overexposed by 2 stops and then recovered in post:

Once again, it is very clear that the sRAW format is a bad choice on the D810 – highlight recovery is very poor when compared to the full size RAW file.

Now, let’s see how the Nikon D850 does in comparison. First, let’s take a look at the same +4 EV shadow recovery with full-size RAW, mRAW and sRAW formats:

Whoa! That’s a night and day difference between the D850 and the D810, isn’t it? Looks like Nikon is finally saving all 12-bits of RAW data. Actually, according to Iliah Borg, these mRAW and sRAW files actually contain Bayer pattern, not YCC data, as it was with the previous sRAW format! This is good news, because those mRAW and sRAW files are going to be much more usable when compared to the bad sRAW file implementation on the D810. They are still limited to 12-bit data, but that’s still a big step in the right direction – something the Nikon D810 should have had in the first place.

Let’s take a look at the highlight recovery potential, again overexposed by 2 EV and then recovered in post:

The same thing, highlight recovery looks as good as the 12-bit original!

Now the next question – does mRAW and sRAW actually reduce image noise at high ISOs when compared to the original RAW image? That’s the result one would expect to see when an image is reduced in resolution. Let’s take a look!

mRAW / sRAW High ISO Noise Performance

It is important to point out that when the below images were down-sampled from full-resolution to matching mRAW / sRAW resolution. As a result, the full RAW images do appear sharper and that’s something we can expect to see when going from a higher resolution image to a lower resolution one. But the biggest question is, does mRAW and sRAW actually have any benefits in terms of high ISO performance? We will start by analyzing high ISO images of full RAW (45.7 MP) vs mRAW (25.5 MP) at specific ISOs like 6,400 and 12,800:

We can see that the images look somewhat comparable in terms of noise performance at ISO 6400. The main difference here is the size of grain – a full RAW image down-sampled to 25.5 MP produces smaller grain in the image, but the overall noise levels look fairly similar between the two. This is another great news for those who decide to shoot mRAW, because it shows that selecting a smaller mRAW format will not result in higher overall noise in the final image. Let’s take a look at ISO 12,800:

Again, we can see similar behavior at ISO 12,800 – aside from the higher amount of sharpness as a result of down-sampling of the full RAW file, there is no added grain in the mRAW image.

Now let’s compare full RAW (45.7 MP) with sRAW (11.3 MP), again at the same ISOs:

The same thing, sRAW shows comparable performance to full RAW that is down-sampled to the same size, which is great. In fact, in some areas of the image, the sRAW image arguably shows less noise, because noise patterns are smoother. What about ISO 12,800?

As expected, the situation does not change much – sRAW performs admirably when compared to the resized full RAW image.

So if you are wondering if you should be shooting mRAW and sRAW formats when you don’t need a full-resolution RAW image, as long as you don’t need 14-bits of data (as pointed out above, mRAW and sRAW are limited to 12-bits), mRAW and sRAW are very good choices to pick, because they don’t increase image noise and provide similar results as a full RAW downsampled/resized in post. So in a way, this does make the Nikon D850 a 3-in-1 camera. For photographers who do different types of photography (such as portraits and landscape), this feature will alone make it worthwhile to upgrade to the D850.

ISO 32 and ISO 64 Benefits

The amazing feature both Nikon D810 and D850 have is the ability to shoot at ISO 64. This very unique capability allows both cameras to capture an incredible amount of dynamic range that can challenge cameras with even larger sensors. To date, no full-frame camera on the market has been able to match the D810 performance at ISO 64, which is a pretty big achievement on behalf of Nikon. And based on what I am seeing so far, the D850 is going to be able to at least match that superb performance, so we should have the new king of dynamic range among full-frame cameras. As I have already shown in other articles and reviews, there is a pretty big difference in dynamic range performance between ISO 64 and ISO 100, so if you shoot landscapes or architecture and you want to be able to preserve as much shadow and highlight information as possible, you should always default to ISO 64. With the added resolution of the D850, this means that shooting hand-held at such low ISO is going to be challenging, especially in less than ideal lighting conditions, so you are going to be better off keeping the D850 on a tripod to reduce chances of camera shake.

If you are not a landscape photographer and you are wondering about the benefits of ISO 64 and 32, keep in mind that those can be very beneficial in a number of situations. First, if you like using fast f/1.4 lenses in broad daylight, you will quickly discover that a neutral density filter is very useful in those situations in order not to overexpose your images. Well, if your camera can lower its ISO to something like 32, you can often avoid having to use a neutral density filter with the shutter speed pushed to 1/8000th of a second. Unfortunately, Nikon continues to keep its maximum shutter speed limited to 1/8000th of a second. That’s understandable for regular shooting, but when using live view and electronic shutter, why not allow shooting faster than that? Fuji and Olympus can shoot at shutter speeds like 1/16000 and 1/32000 utilizing electronic shutter, so Nikon should also add that as an option. That would be a powerful capability to have, especially when combined with a lower base ISO.

Second, lower ISO can be useful when photographing long exposures during the day, say moving people, water, etc. Sure, you can always stop down your lens to a smaller aperture, but at the expense of diffraction. Anything past f/8 on a high-resolution sensor negatively impacts image quality. If you compare the D850 with the older D700, there is a 1.66 stop difference between the lowest allowed ISO setting on both – a significant difference. If your starting shutter speed at say f/8 is 1/250 of a second at ISO 100, going down to ISO 64 will reduce the shutter speed to 1/160 and going down to ISO 32 will reduce it to 1/80. That’s almost equivalent to using a two-stop ND filter in front of the lens. It is nice to have these options when working in the field. The difference is obviously diminished to 2/3 of a stop when you compare the D850 with the newer cameras that have a base ISO of 100, but it is still an advantage. Cameras usually struggle with highlight recovery a lot more than shadow recovery, so that 2/3 of a stop advantage can sometimes save the day.

Third, flash photography fans will also appreciate the ability to lower ISO when using fast lenses and wanting to isolate their subjects from the background. Since flash sync is limited to 1/250 of a second, the shutter speed cannot be pushed beyond that, or part of the frame will be rendered dark. And 1/250 in broad daylight means that one would have to stop the lens down to a smaller aperture like f/5.6 or smaller in order to avoid overexposure. Well, f/5.6 is a pretty small aperture for a fast prime lens – it obviously translates to a larger depth of field, which is not always desirable.

What if you want to have a nice portrait, with the subject isolated from the background? Or perhaps you want to make those Christmas lights in the background appear like blobs with nice bokeh characteristics? Well, the only way you can do that without altering ISO is by using a neutral density filter. By blocking some of the light, you can essentially open up the lens more and decrease depth of field, which means more of the background appearing out of focus. If we look at the same 1.66 stop difference example mentioned above between Nikon D700 and Nikon D850, by changing from ISO 200 to ISO 64, you would be able to maintain the shutter speed at 1/250 of a second, while being able to open up the lens from f/5.6 to f/3.2, which is a huge change in depth of field. And sometimes even 2/3 of a stop of exposure makes a difference in darkening the sky a little when overpowering ambient light.