Sensor Size Vs Noise Comparison

If you would like to post, you'll need to register. Note that if you have a BCG store account, you'll need a new, separate account here (we keep the two sites separate for security purposes).

Don'sPhotoStream

Member
Supporting Member
Marketplace
I continually read that a 40+ mp camera is far better because it lets you shoot bird and other small wildlife photos and then crop in post to get an acceptable subject size in the frame.

Yet, my experience has been (with all the D800 series cameras) that I can achieve better results with the 24 mp cameras, i.e., D750, D5, D6, and Z6, because of the lower low-light noise-level of these cameras.

My question is this, does the Z7 change the game and allow you to crop significantly in post, say 100% crop and craft a more noise-free image by down sampling the image, than the Z6 under the same conditions. Does anyone have experience with this situation?

I really like my Z6 ll (except for the focus system) because the image stabilization brings new life to my existing lenses, such as the 80-400/5.6. So, I'm considering the purchase of the Z7 ll.

I rarely print larger than 11x14; 99% of my work is for posting on Flickr, Instagram, etc.

Thanks for any feedback that is offered.
 
Cropping and upsizing itself magnifies noise and thus reduces dynamic range. Downsizing reduces noise and improves dynamic range. I think what matters is how much light was gathered per each image area in the final image. These two resources might be partial answers:


 
Noise is more apparent with great magnification of any image. I test any new camera at ISO 6400 with a tungsten light source and reds and blues in subjects that make it easy to see if chroma noise will be a problem. Surprising to me it has not been a problem with the D500 or D850 cameras. I also differentiate between noise that is readily corrected in auto NR with an application and a noise level that may take a good deal of time to correct without significant loss of detail.

There are many variables involved in getting signals from the sensor processed in the camera and the creation of the Raw output file. The D850 with its 45MP sensor shows less noise than I saw with the D3 and its 12MP sensor so technology has improved considerably over the years. The D850 can provide a APS-C 19MP file as compared to the D5/D6 that produce a 8.9MP crop so a great deal more image magnification may be needed with these 20MP cameras.

I have also done experimentation with shooting at +1 EV and -1 EV as well as at 0.0 EV to see in which situations it was better to overexpose or underexpose and then make adjustments in post. With the D3 for example a 1 stop overexposure would produce better IQ in post than a 1 stop underexposure. Recommend experimenting along these lines as it takes very little time to do this and then decide for yourself which is better. Part of what I have seen is color shifts and a loss of color fidelity with underexposures with the cameras I have used.
 
I find that DX crop on a D850 or Z7 looks, for most intents and purposes, like a frame from D500 or Z50. In theory a D780 or Z6 would have better noise characteristics in shadows and low light, but I'm not at all sure the difference is photographically that significant. But you can certainly be more aggressive in cropping the 45mp bodies. Lots of factors, lots of choices.
 
A couple of reviews show that the 61mp Sony A7R IV loses its resolution advantage over the 42mp R III to noise once you get over 400 ISO.
You can reduce grain in post but you don't get the detail back.
 
It also depends on how flexible your setup needs to be. I like the D850 because it is superbly adaptable, from high resolution landscape at low iso on tripod, to half resolution crops At higher iso for wildlife. High iso is not as good as a D5 but up to 4000 I can’t tell the difference (on noise) with my D750. At 6400 the D750 has a tiny edge and I don’t shoot the D850 any higher. But in the end it is very good at everything and excels at a few. A D5 is a better wildlife camera if you have the reach (big if, as in 600 f:4 big) but it won’t touch a 45 mp landscape shot at iso 64 on tripod.
if I had a dedicated camera for wildlife with a 600mm f:4, I would have the superior D5 or D6 for AF and iso but since I don’t, the most flexible gear wins the day.

In the end, it is all about compromise based on your priorities.
 
I think one can make too much of the noise concerns for most uses. Below a certain point (ISO 1600?) noise is pretty negligible from most modern sensors, and even above that it's most visible in boosted shadows. And up to a certain point noise is manageable in post processing software. Cropping flexibility is certainly an issue, but that depends a lot on what you're shooting and how close you are (those damned warblers). Add to that the various sensor shift, high res features, and stitching options and you end up with dozens of variables on how to get good sharp images. I shoot 18mp Nikon V3, 20mp Olympus OM-D, 20mp D500 and Z50, 24mp D3x, and 45mp D850 and Z7 and find them all manageable and good for different tasks. I'm not as able to crop in with a V3, but I'm likely to be getting places I can't with full-frame, and effectively closer (810mm). By now I think you can get good 12x18 prints or web images from almost anything on the market (I know people who swear by their Nikon P900s). (yes, I have too many cameras) :)
 
It's not just about resolution, but the relationship between resolution (sensil size), Dynamic Range, and noise. There are professional portrait photographers who still use D3s with a 12.3mp sensor because they can always fill the frame. I put the FX body away and pull out the DX body when I can't fill the FX frame in order to get more pixels, read that as the ability to capture more detail, on the subject. So it's not the sensor resolution exactly, but the number of pixels you can get on the subject. Most modern sensors have similar capability. This chart is from DxOMark.com. Look at how close the D850 D500 and D800 track eachother in terms of DR and noise (noise is the bar graph on the right). When you crank up the ISO DR falls quickly and noise goes up quickly with all these sensors. The Nikon Z sensors are no better than the D850 and D500 sensors.

So the point is, no matter what your camera, always try and get as many sensor pixels on your subject as you can using the best lens you can afford for the best possible detail and keep your ISO as close to base as you can for greatest dynamic range and lowest noise. Try and pick a camera that allows you to do this for the type of photography you practice. All sensors degrade similarly in performance as ISO goes up.

D850 800 500  DR.JPG
You can only see EXIF info for this image if you are logged in.
 
There isa good reason why DxO is looking at the four criteria
  • Signal Noise Ratio
  • Dynamic Range
  • Tonal Range and
  • Color Sensitivity
If you imagine these criteria being tetrahedron, the pivot point for what you want to do is somewhere between these four edges and implicitely what want to doand how you want to do it. The tetrahedron get reshaped depending on which equipment (camera) you use and which ISO you need to shoot with, where the most significant influence is created by ISO because it basically makes the tetrahedron smaller.

Although the values all four values decline with growing ISO values, because that's basically the physics of optical sensors. There are considerable differences to consider.

The second thing that they did for good reason is distinguishing measured vakue depending on the media you use for displaying the resulting photos. In many cases the differences between cameras / sensors are much bigger for active display, i.e. projection by menas of screens or beamers, than they are for passive display, i.e. print. I have seen pictures looking great as a large format print but were pretty much unusable for large scale display with a high res beamer.

IMHO for images that are downsampled and converted to sRGB for internet publications more or less any shot that is properly taken is good enough if looking at the modern DLSR generations with the exception of publishing hi res images or even native res images for download.

Because I wanted to get a good decision whether to replace my D750 and D7200 by one D850 or not and whether or not to keep my D4s for low light or not, I did some research along this line. Due to the fact that "you can tweak data by means of software, but you can't cheat physics", I prefer to avoid or minimze noise in the pictures as early as possible and this is at the time you take the image. Yes, I know and I love things like DxO Deep Prime, but I still prefer to use it only for stretching the limits if needed in cases where I can't reduce the noise to an acceptable level upfront, rather make it a default.

Keeping in mind that one of the key criteria from physical perspective is the size of a sensor receiving light and transforming it into an electrical signal - and fporm there into data - we need to look at the size of an individual pixel, because it is the one causing the noise.

I was comparing in two directions, one being the capability of the D850 replacing my D750 and D7200 and the second being the question of whether I can get any better regarding IQ with really high ISO. Leaving the fist part aside in this context here, I looked at my D4s, the D5 and the D850 (I would have taken the D6 but they don't have that in the database yet). Here are the charts:

SNR18.jpg
You can only see EXIF info for this image if you are logged in.


From the points where the camera lines cross the dotted red line you can see that the D5 is about 1 stop better compared to the D850 and the D4s another 0.7 to 1 stop better than the D5.

Regarding the dynamic range D4s and D5 pretty much on par from ISP 1600 onwards while the D850 is behind for high ISO, sometimes up to 1.5 stops.
Similar observation is made for the Tonal Range as well as the Color sensitivity, exceapt that thecurves are not that mixed up towards low IOS values.

DynRange.jpg
You can only see EXIF info for this image if you are logged in.


TonalRange.jpg
You can only see EXIF info for this image if you are logged in.


ColorSense.jpg
You can only see EXIF info for this image if you are logged in.


What does that mean keeping in mind the pixel size thing ?
  • If you stay more or less within one generation of a sensor technology you can compensate part of the disadvantages of smaller pixels by means of more advanced in-camera software, but the smaller pixel can't substantially ovetake it s bigger counterpart.
    In terms of sensor performance the D5 can catch up to the D4s in three of four discipline despite having higher resolution and thus smaller pixels and in the SNR discipline it stays behind its older counterpart, whose pixels are about 1/3 bigger.
    Of course there are other criteria like AF speed, precision and accuracy in low light and there the D5 certainly wins, but from the sensor perspective there would be no point for upgrading.

  • The D850 is for sure one of the best - if not the best - DSLR that Nikon has built so far, but as physics tells us, it has to pay a price for its monster resolution and the currency in which it is payed is pixel size. In three disciplines it is one stop behind the big single digit machines and it terms of noise it is 1.7 stops compared to the D4s.
    That said, don't get me wrong here: I loved the D850 from the moment I took the first test shots after taking it out of the parcel for its IQ and versatility !
The conclusion was
  • to keep the D4s for the situations where light gets really critical,
  • to get the D850 for everything else and replace my D750 and D7200,
  • keeping the chance to have to high end bodies usable in parallel if required including having the additional reach I used to have with my D7200, because the D850 provides already in DX mode about 20% more resolution than the D4S and only about 6% less compared to the D5
Again and always it all depends on you, what you do and how you do it. In terms of sensor capabilities it comes down to getting the pivot point of your work inside the tetrahedon defined by your camera and only if you find you can't get it inside you need to think about changing somthing.
 
There isa good reason why DxO is looking at the four criteria
  • Signal Noise Ratio
  • Dynamic Range
  • Tonal Range and
  • Color Sensitivity
If you imagine these criteria being tetrahedron, the pivot point for what you want to do is somewhere between these four edges and implicitely what want to doand how you want to do it. The tetrahedron get reshaped depending on which equipment (camera) you use and which ISO you need to shoot with, where the most significant influence is created by ISO because it basically makes the tetrahedron smaller.

Although the values all four values decline with growing ISO values, because that's basically the physics of optical sensors. There are considerable differences to consider.

The second thing that they did for good reason is distinguishing measured vakue depending on the media you use for displaying the resulting photos. In many cases the differences between cameras / sensors are much bigger for active display, i.e. projection by menas of screens or beamers, than they are for passive display, i.e. print. I have seen pictures looking great as a large format print but were pretty much unusable for large scale display with a high res beamer.

IMHO for images that are downsampled and converted to sRGB for internet publications more or less any shot that is properly taken is good enough if looking at the modern DLSR generations with the exception of publishing hi res images or even native res images for download.

Because I wanted to get a good decision whether to replace my D750 and D7200 by one D850 or not and whether or not to keep my D4s for low light or not, I did some research along this line. Due to the fact that "you can tweak data by means of software, but you can't cheat physics", I prefer to avoid or minimze noise in the pictures as early as possible and this is at the time you take the image. Yes, I know and I love things like DxO Deep Prime, but I still prefer to use it only for stretching the limits if needed in cases where I can't reduce the noise to an acceptable level upfront, rather make it a default.

Keeping in mind that one of the key criteria from physical perspective is the size of a sensor receiving light and transforming it into an electrical signal - and fporm there into data - we need to look at the size of an individual pixel, because it is the one causing the noise.

I was comparing in two directions, one being the capability of the D850 replacing my D750 and D7200 and the second being the question of whether I can get any better regarding IQ with really high ISO. Leaving the fist part aside in this context here, I looked at my D4s, the D5 and the D850 (I would have taken the D6 but they don't have that in the database yet). Here are the charts:

View attachment 15196

From the points where the camera lines cross the dotted red line you can see that the D5 is about 1 stop better compared to the D850 and the D4s another 0.7 to 1 stop better than the D5.

Regarding the dynamic range D4s and D5 pretty much on par from ISP 1600 onwards while the D850 is behind for high ISO, sometimes up to 1.5 stops.
Similar observation is made for the Tonal Range as well as the Color sensitivity, exceapt that thecurves are not that mixed up towards low IOS values.

View attachment 15197

View attachment 15198

View attachment 15199

What does that mean keeping in mind the pixel size thing ?
  • If you stay more or less within one generation of a sensor technology you can compensate part of the disadvantages of smaller pixels by means of more advanced in-camera software, but the smaller pixel can't substantially ovetake it s bigger counterpart.
    In terms of sensor performance the D5 can catch up to the D4s in three of four discipline despite having higher resolution and thus smaller pixels and in the SNR discipline it stays behind its older counterpart, whose pixels are about 1/3 bigger.
    Of course there are other criteria like AF speed, precision and accuracy in low light and there the D5 certainly wins, but from the sensor perspective there would be no point for upgrading.

  • The D850 is for sure one of the best - if not the best - DSLR that Nikon has built so far, but as physics tells us, it has to pay a price for its monster resolution and the currency in which it is payed is pixel size. In three disciplines it is one stop behind the big single digit machines and it terms of noise it is 1.7 stops compared to the D4s.
    That said, don't get me wrong here: I loved the D850 from the moment I took the first test shots after taking it out of the parcel for its IQ and versatility !
The conclusion was
  • to keep the D4s for the situations where light gets really critical,
  • to get the D850 for everything else and replace my D750 and D7200,
  • keeping the chance to have to high end bodies usable in parallel if required including having the additional reach I used to have with my D7200, because the D850 provides already in DX mode about 20% more resolution than the D4S and only about 6% less compared to the D5
Again and always it all depends on you, what you do and how you do it. In terms of sensor capabilities it comes down to getting the pivot point of your work inside the tetrahedon defined by your camera and only if you find you can't get it inside you need to think about changing somthing.

One thing I read about dxo is it does not normalize for image size, which can lead to a different result when comparing sensors of different size. Photonstophotos.net on the other hand does normalize, so the difference is not so great in that case.

 
Last edited:
There isa good reason why DxO is looking at the four criteria
  • Signal Noise Ratio
  • Dynamic Range
  • Tonal Range and
  • Color Sensitivity

I was comparing in two directions, one being the capability of the D850 replacing my D750 and D7200 and the second being the question of whether I can get any better regarding IQ with really high ISO. Leaving the fist part aside in this context here, I looked at my D4s, the D5 and the D850 (I would have taken the D6 but they don't have that in the database yet). Here are the charts:



From the points where the camera lines cross the dotted red line you can see that the D5 is about 1 stop better compared to the D850 and the D4s another 0.7 to 1 stop better than the D5.

Regarding the dynamic range D4s and D5 pretty much on par from ISP 1600 onwards while the D850 is behind for high ISO, sometimes up to 1.5 stops.
Similar observation is made for the Tonal Range as well as the Color sensitivity, exceapt that thecurves are not that mixed up towards low IOS values.


What does that mean keeping in mind the pixel size thing ?
  • If you stay more or less within one generation of a sensor technology you can compensate part of the disadvantages of smaller pixels by means of more advanced in-camera software, but the smaller pixel can't substantially ovetake it s bigger counterpart.
    In terms of sensor performance the D5 can catch up to the D4s in three of four discipline despite having higher resolution and thus smaller pixels and in the SNR discipline it stays behind its older counterpart, whose pixels are about 1/3 bigger.
    Of course there are other criteria like AF speed, precision and accuracy in low light and there the D5 certainly wins, but from the sensor perspective there would be no point for upgrading.

  • The D850 is for sure one of the best - if not the best - DSLR that Nikon has built so far, but as physics tells us, it has to pay a price for its monster resolution and the currency in which it is payed is pixel size. In three disciplines it is one stop behind the big single digit machines and it terms of noise it is 1.7 stops compared to the D4s.
    That said, don't get me wrong here: I loved the D850 from the moment I took the first test shots after taking it out of the parcel for its IQ and versatility !
The conclusion was
  • to keep the D4s for the situations where light gets really critical,
  • to get the D850 for everything else and replace my D750 and D7200,
  • keeping the chance to have to high end bodies usable in parallel if required including having the additional reach I used to have with my D7200, because the D850 provides already in DX mode about 20% more resolution than the D4S and only about 6% less compared to the D5
Again and always it all depends on you, what you do and how you do it. In terms of sensor capabilities it comes down to getting the pivot point of your work inside the tetrahedon defined by your camera and only if you find you can't get it inside you need to think about changing somthing.

Your analysis is interesting and well done, but I think you are using the wrong metric.

The resolution of the sensors vary widely and a proper comparison normalizes the values. In DXO speak this is done with the print mode. Here are the results with normalization:

dxo.jpg
You can only see EXIF info for this image if you are logged in.


The conclusions are different and the D850 looks much better.

Cheers,

Bill
 
Last edited:
A D750 with a DX amount of crop is effectively a 13MP camera. If you have a full frame image with no need to crop then noise may be a consideration but if you need to enlarge the image by 50% then the gain in lower perceived noise is likely to be lost.

Having worked with Nikon cameras with terrible ISO performance like the D2x that was usable only up to ISO 640, I have learned to differentiate between different types of noise and to determine with my own testing how high an ISO setting will still result in usable images. At a given ISO the noise may be eliminated for all intents and purposes with an Auto NR application and at a higher ISO level it may still be rescued with manual adjustments. The D3 for examply was fine up to ISO 2400 but I could get usable images at ISO 4000 with 95% certainty and at ISO 6400 with 70% certainty. I found that with my D7200 the noise was on par at ISO 6400 with the D3 at ISO 2400 which I found quite remarkable.

It is worth the effort to do your own testing and to evaluate noise and also whether with a given camera and its senor it is better to over expose or under expose in situation where there is wide DR in the scene. With the D3 for example, I found that overexposed images could be recovered to a far greater degree than underexposed images in terms of color fidelity and tonal range. But it is camera specific as the technology changes a great deal year to year.
 
Your analysis is interesting and well done, but I think you are using the wrong metric.

The resolution of the sensors vary widely and a proper comparison normalizes the values. In DXO speak this is done with the print mode. Here are the results with normalization:

The conclusions are different and the D850 looks much better.

I can understand your arguments, but I don't quite agree and here's why:

The comparison labelled "print" at DxO looks a standardized output generated from an original image, in this case a standardized print.
Standardizing prints of images created by cameras with different resolutions in terms of size implicitely includes downsampling of the higher res image.
It is known that prints are much less prone to visible noise compared to looking at images that are displayed actively and thus is logical that the graphs for the different cameras get closer together on the "print" page. But if you work primarily with displaying images this comparison is useless and this is exactly the reason why DxO distinguishes between "screen" and "print".

The other thing is: Why on earth should I buy one of these resolution monsters if the first thing to do is downsampling the image to reduce the noise ???
If I get in the situation that I am not able to fill the frame due to distance to the object or missing reach in the optics and I want to utilize the high resolution by means of heavier cropping, that I usually try to avoid for obvious reasons - being it in post or in-camera by shooting the D850 in DX mode, the only thing that interests me is the native level of noise I see on the display, not what it might look like if I print it.

I agree that my perspective is not right for everyone, but for those looking for high quality in actively displayed hih res images and not having the priority on printing this should be the way to look at it ...
 
Last edited:
It's a very interesting discussion. But what's the final conclusion ? If I shoot the same scene with the Z6 and the Z7, and crop the Z7 down to 24MP, using the same iso, exposure and shutter, will I get noticeably more noise with the Z7 ?

And would it be fair to say that the D500 which is now 1 sensor generation behind the Z7 II will still be comparable with it ? Obviously it has the better autofocus for wildlife, and also smaller pixels so it would be interesting to know given pixel size, how these two compare.

Thanks for any answers.
 
I believe the Z7 image will show more noise, in the conditions you state.

Hope someone proves me wrong. And I’m talking bird photography where I’m usually cropping 100 percent.
 
Yes, for sure it must be worse. That's physics. But how much worse is the question isn't it ? You'd live with some noise for flexibility to crop down, but not too much. Hope someone has done a comparison like I described and can upload samples for us to see.

I googled a bit about the D500 sensor and compared to the Z7 and I found that the Z7 pixels are actually larger, so I think a Z7 full frame BSI sensor is going to be better for noise (and other things because its also a generation newer) than the D500 CMOS sensor. But I don't think the Z7 autofocus is there yet, so maybe it makes more sense to go to a D850 instead if you are moving from DX to FX and need to buy now. (pixel pitch of 4.35μm for Z7 versus 4.22μm for the D500)

If you are already on mirrorless and do low light often, then you may want to stick to Z6 rather than Z7 and crop, unless the difference is minimal and the flexibility you get with the ability to crop is therefore the advantage.
 
It's a very interesting discussion. But what's the final conclusion ? If I shoot the same scene with the Z6 and the Z7, and crop the Z7 down to 24MP, using the same iso, exposure and shutter, will I get noticeably more noise with the Z7 ?

And would it be fair to say that the D500 which is now 1 sensor generation behind the Z7 II will still be comparable with it ? Obviously it has the better autofocus for wildlife, and also smaller pixels so it would be interesting to know given pixel size, how these two compare.

Thanks for any answers.
I believe the Z7 image will show more noise, in the conditions you state.

Hope someone proves me wrong. And I’m talking bird photography where I’m usually cropping 100 percent.

Although I don't own a Z camera, I think physics is telling the answer:

Providing that the pixels are more or less identical from the technological side - apart from resolution of course - you most likely will see more noise in the cropped Z7 image, simply because you have roughly twice the resolution for the same sensor size (FX) and thus the sensor area per pixel is roughly half of what you have in the Z6.

Looking at it from a different perspective, if you look at what @Steve wrote / told us about noise in relation to ISO and cropping:
The more you crop the more visible the noise becomes visible because by cropping you are "magnifying" the object AND the noise with it.
As downsampling an image usually helps to reduce the visibility of noise I would expect to see similar results between the two models if you downsample the Z7 image to Z6 resolution, but this is something that should be answered finally by someone here who has the two cameras to test it.

That said, I can't tell how relevant it is to you because I don't have personal "noise experience" ;) with a Z6 or Z7.
What I can say, that this principle proved to be true for my DSLRs and comparing the noise levels depending on the sensor pixe size there between cameras that are within one Expeed generation of sensors, i.e. you don't compare sensors that are two far apart from the technological "age".
 
Last edited:
Back
Top