Nikon Cameras Noise Comparison—D700, D800, D3X, D3S, and D4

My friend Drew Eschbacher says I haven’t mastered the concept of the three-paragraph blog post and as a result lose some readers along the way. He’s right, and this post will only reconfirm his assessment. You are welcome to scroll down to the end for the conclusion, but you just might miss something important if you do. At the least, you’ll miss some things I’d really like for you to know.

I realize that this comparison will be a bit late for some of you. After all, the Nikon D4 and D800 have been out for a whole whopping 6 months now. In the social media world of new cameras from Nikon and Canon, they’re ancient history. Unfortunately, I couldn’t do anything until now because I couldn’t afford either of the cameras. Still can’t, but I do have the good fortune of having friends who bought them and were willing to loan them for testing. It just took until now to gather them up from everyone.

I currently shoot with the Nikon D700. The primary goal with this test was to determine which of the current full-frame (FX) cameras from Nikon produces the least amount of noise so I would know where the D700 stands in this lineup. I’m a night photographer. Shoot most photos in low, or NO light. Use high ISOs and long exposures. Noise is a big deal.

Of course, low noise is only one component of a good photo. Dynamic range, color fidelity, and a host of other factors come into play. However, as far as I’m concerned, ALL current professional cameras from Nikon and Canon are close enough in these departments that I don’t need to consider them when making a buying decision. Sure, there are differences, but they are subtle. Not so with noise. For a night photographer, it’s the deal breaker.

Scientifically defining the noise characteristics of a camera is far beyond my capabilities and even farther beyond any desire to do so. I couldn’t give a flying flip about the math behind noise and the technical definitions of the many types of noise. I care about what my photos look like. That said, I knew that there was no way to devise a real-world test for all of these cameras. For one thing, I wasn’t able to test them all at the same time, so there’s no way I could shoot the same outdoor subjects in the same lighting. But that’s okay; in a test for only noise, you don’t have to shoot real subjects. All I cared about was COMPARING the cameras, so I just needed to make sure to be consistent. Make sure they all played by the same rules.

All exposures were RAW (Nikon NEF) with Long Exposure Noise Reduction turned off and High ISO Noise Reduction set to Normal. Exposures were made in a controlled environment, in temperatures averaging between 65 and 75 degrees. The one parameter that wasn’t consistent is that I didn’t shoot all exposures from start to finish, without interruption. It took so long to perform a complete cycle for each camera that I had to resume some shots the following day. In doing so, the camera cooled overnight and the remaining exposures likely exhibited less noise than they would have otherwise. (The amount of noise is directly related to temperature, with more noise at higher temps.) However, as best as I can tell, this did not grossly affect the results and it had very little bearing on the COMPARISON between cameras.

All exposures were shot with the lens cap on or with no lens and a camera cap in place. A black cloth covered the camera in all exposures. Since no light reached the sensor, the only thing it recorded was the noise. Yes, I’m fully aware that this tells nothing about how the noise appears in a real image, particularly after it goes through post processing. But remember, the purpose here was simply to compare the cameras. Photographing real-world scenes would have introduced factors that would influence this comparison. Granted, the noise characteristics of a particular camera might appear different in a real photo than they do in the test shots, but the comparison between cameras is accurate.

Why only full-frame Nikons? Why not, for instance, the excellent D300 or D7000? Admittedly, I allowed a few preconceived notions to affect the decision. See, the generally believed consensus is that with ALL ELSE EQUAL, the smaller the sensor, the more noise you get. Or, more specifically, the larger the pixel size, the less noise. A larger sensor doesn’t have to cram tiny pixels onto it as much as a smaller sensor does. I had already tested the D700 against the D300 a few years ago and saw that the FX sensor of the D700 produced less noise than the DX sensor of the D300. I naturally assumed that ANY current FX sensor would be superior to a DX sensor with regard to noise. And in any comparison of FX sensors, the ones with the largest pixels (fewest megapixel count), would naturally be assumed to produce the least noise.

After reviewing the results of this test, I’ve kicked that assumption to the curb!

The D700 is 12.1 megapixels, the D3S is 12.1, the D4 is 16.2, the D3X is 24.5, and the D800 is 36.3. If pixel size were the main determining factor with noise, the D700 and D3S should be the winners and the D800 the clear loser, right? Actually, except for the D3S, it’s closer to the reverse!

The thing is, all else is NOT equal. Pixel dimension is only one of many factors, not the least of which is the processing engine behind them. I don’t know exactly how a sensor works and I frankly don’t care. Understanding the physics behind digital cameras does nothing to improve my photography. As I’ve learned over the years and as this test has reconfirmed, simply understanding how a camera works will not necessarily allow you to make the best decisions about using them. For that, you need to step away from lab and take some pictures. That said, I stand firm in the belief that the more you know about the mechanics of photography, the easier it is to concentrate on the creative aspects.

All of the cameras tested are in Nikon’s current lineup except for the D3S, although it may still be available new from some dealers. I included it because until the release of the D4, it has been generally accepted as the de facto leader among low-noise cameras, both from Nikon and Canon, and I figured it would make a good benchmark. Some of the social media buzz suggests that the D4 has surpassed the D3S in the noise department, but this test does not support that conclusion.

Random noise generated by the Nikon D700

This image shows considerable random noise. This type of noise is not consistant, so it is not easily removed by the use of software without degrading the image.


Non-random noise generated by the Nikon D4

This image shows considerable non-random noise, represented as colored specs scattered all over the image. This type of noise is consistant, so it is easily removed by the use of dark frames, such as when using long-exposure noise reduction.

There are several different types of noise and other undesired effects that occur with high ISOs and long exposures and a lot of math behind all of them. I’d rather snuggle with an alligator suffering from a toothache than to try to wrap my head around the science behind it all. For this test, I was mainly interested in the two general types of noise: random and non-random.

Random noise is what most people think of when they think about noisy images. It’s roughly the equivalent of the grain we got with high-speed films. Random noise consists of several different types of specific noise all caused by different inefficiencies or physical characteristics within the digital imaging system. Generally, random noise becomes greater as the ISO increases. Because it is not the same from one exposure to the next—thus, random—it is not easy to remove using software without softening the image.

Non-random (or fixed-pattern) noise consists of what are called stuck, hot, or dead pixels—tiny colored or black specs scattered all over the image. This type of noise generally increases with longer exposure lengths. Because non-random noise is consistent, it is easily removed in software by the use of dark frames. When you use Long-Exposure Noise Reduction in your camera, the camera takes a dark frame and uses it to remove the fixed noise.

I’m afraid there’s no practical way to show you all the test exposures, as they total nearly 9 gigs. Converting them to JPEGs and downsizing would alter the noise characteristics, making an accurate evaluation impossible. You’ll just have to take my word for it, or else perform your own test.

Click here to open the test chart.

Reading the test chart should be intuitive except for the rating system. I viewed all the RAW files at 100 percent and judged the noise on a scale of 1 to 5, with 1 between no discernible noise and 5 being excessive noise. Each frame has two numbers. The first one represents random noise, while the second number is for non-random noise. Since non-random noise is much easier to remove during processing (or capture, with long-exposure noise reduction) than random noise, the first number is much more important in an evaluation of the results. Anything above a 3 means the image is pretty much unusable, although in certain situations it might be salvageable.

Okay, I know what you’re thinking, so go ahead and say it. “Kevin, you’re applying a subjective evaluation to a mostly objective test.” Yep, it’s true. But I have no way of objectively measuring the noise in these files, so I simply looked at them and picked a number. I viewed each set of ISO/shutter speed series side by side, so at least those are reasonably objective. For instance, it is reasonable to assume that the comparison between all the photos shot at ISO 200 and 15 minutes is fairly accurate, but you cannot make the same assumption if comparing the results between different ISOs and shutter speeds because once I closed a particular file, I could no longer remember its qualities accurately enough to compare it to others. So, you have to look at the big picture and view the results in the chart as a trend.

An interesting discovery is that in a few cases the image quality improved as shutter speed increased. For instance, with the D3S at ISO 12,800, photos shot at 1/2 second exhibited less noise than those shot at 1/30 second. If I were shooting real scenes, I would attribute this to a better signal-to-noise ratio at 1/2 second. But I wasn’t recording any light at all, only noise, so I am at a loss to explain it. Perhaps this has something to do with readout noise from the processor.


If you don’t care to wade through the test chart, here’s a general evaluation of the results.

Best all around: Without question, it’s the D3S. If you primarily shoot long exposures and/or high ISOs and you can stomach $5K for a used camera, grab one of these puppies while they are still available. The D3S performance frankly shocked me. It is freakin awesome!

Worst all around: Another shocker. It’s a tie between the D700 and D3X. I expected the 12.1 megapixel D700 to be much better than the 24.5 megapixel D3X, but no go. They are virtually neck and neck in the bad department. Bad compared to the other cameras in this test, that is. I certainly wouldn’t have been shooting for years with a camera that I considered bad overall.

Best at high ISOs: Yet another shocker. I expected the D4 to rule in this department, considering that it was manufactured and is marketed for high-ISO performance. However, the D3S slightly outperforms it. At ISOs of 51,200 and above, both cameras produce horrible images, but at ISO 25,600, they produce acceptable photos at shutter speeds up to a couple of seconds.

Best at long exposures:  The D3S is the winner, although at low ISOs the D800 is close. At high ISOs, the D4 takes over the baton and starts nipping at the D3S’s heels.

Biggest surprise: The performance of the D800. Wow! Up to ISO 12,800, the D800 is equal to the D4, and even better in long exposures at low ISOs. Certainly, the D4 is an impressive camera, but the results of this test show that it is no better than the D800 for low-light photography. In only one area does the D4 outperform the D800—with exposures up to about 8 seconds using ISOs of 12,800 or higher. And we’re talking about viewing the D800 files at their native resolution. I’ve read many reports about people downsizing D800 files to match the D700 or other cameras in order to decrease the noise, but this never really made sense to me. And now, I can see that it isn’t necessary, as the native D800 files look cleaner than the D700 files to begin with. 

Which camera should a night photographer own?

That’s not easy to answer, as there are so many other factors to consider besides just the noise. In an ideal world, you’d own the D3S for all-around night shooting, the D4 for high-ISO shooting, and the D800 for when you need a lot of megapixels for large prints or whatever. Unfortunately, I live in the real world, where owning three cameras costing a total of $15K is a fantasy.

My choice for the best all-around camera for both night photography and general shooting is the D800. The D3S has less noise, but it costs a couple grand more and its video functions are not nearly as good as the D800. Video is an important consideration for me. I don’t need the high frames-per-second of the D4, and I darn sure don’t need the extra $3K price tag of it over the D800. The D3X was never a serious contender due to its high price (over $7K) and assumed bad performance in low light, an assumption confirmed by this test. I expected the D700 to perform better than it did and to hold its own as a second camera when I could afford to get a new model, but it disappointed me. I’ll be getting a D800 as soon as possible, after which the D700 will be relegated to daytime shooting and as a backup night camera.

Remember, the primary goal with this test was to help me make the most informed purchasing decision. I share the results with anyone who might make use of them. I understand that some of you might not like the methodology and I expect some will disagree with the conclusions. That’s okay. The test worked well for me.

Now, it’s time to step away from the lab and take some pictures!

Tags: , , , , , , , , ,

Leave a Reply

You must be logged in to post a comment.