Z9 Noise Above ISO 4000 - Is It As Bad As I Have Heard?

chambeshi said:

The established knowledge is that photosite area correlates inversely with SNR is established on the differences in captured photons.

https://www.lensrentals.com/blog/2012/02/sensor-size-matters-part-2/

A published analysis comparing image sensor specifications supports the principle that "...Decreasing pixel size increases the likelihood that photon noise will become visible....".

Catrysse, Peter B., and Brian A. Wandell. "Roadmap for CMOS image sensors: Moore meets Planck and Sommerfeld." Digital Photography. Vol. 5678. SPIE, 2005.

It follows that the effects of higher SNR from larger photosites scales upwards to influence overall sensor noise as a function of pixel count. Compared to smaller photosites (eg 45mp sensor) the advantage of large photosites becomes evident in less sensor noise as ISO increases - eg in low light conditions. In contrast, smaller pixels generate proportionally more noise at higher ISOs. At higher ISO > 6400, the larger photosites of the D6 (basically = D5) 20mp sensor confer a distinct advantage, which Nikon optimized for lowlight photography.

https://www.imaging-resource.com/ne...-inside-nikons-super-secret-sensor-design-lab

The details of the D5 sensor architecture are propriety albeit the tech teardown with Xray data etc is available behind a paywall. But they presumably entail the optimization of microlenses and colour filters etc (as much as Etchels was permitted by Nikon to see for his IR article). There must be sound reasons why Nikon designed the D5 sensor on the foundation of 6.41 µm photosites and not on a D800 spec sensor (4.88 μm) or even smaller photosites. It follows the sensor engineers exploited the ability of these larger photosites to capture more light with higher SNR at higher ISOs, and confer a higher SNR on the lowlight advantage of the 20mp sensor.

The applications of these advantages in lowlight imaging extend beyond terrestrial photography into cutting edge scientific applications. It appears NASA scientists did their homework carefully in selecting the Nikon D5 (not the D800 sensor) for the ISS:

https://www.sciencedirect.com/science/article/pii/S003442572100331X

Notwithstanding that we can presume Nikon designed the Z9 BSI stacked sensor several years after that in the D5, the evidence is clear in the comparative sensor data (eg Photonstophotos) and Z9 images at higher ISOs that optimization of sensor micro architecture faces finite limits to mitigate the constraints of smaller photosite on noise.

Click to expand...

chambeshi said:

FingerPainter said:

No. For it to be "knowledge", it has to be true, and it isn't. It has been a common mistaken belief of many misinformed people for many years now. Photosite size has very little to do with the SNR performance of whole sensors. it has nothing to do with the shot noise of a sensor. A sensor captures the same number of photons at a given exposure regardless of how many photosites it is divided into.

People who mistakenly believe that larger photosites lead to less noisy images mostly don't understand that the noisiness of the image is not the same as the noisiness of an individual pixel. They mistakenly think that since a larger photosite captures more light, and therefore has a higher SNR, that the whole image must therefore have a higher SNR. They forget to account for the fact that there are more of the smaller photosites on a sensor of the same size, and this extra number of photosites exactly makes up for the smaller size of the photosites. A sensor of a given size captures the same number of photons at a given exposure regardless of the size of its individual photosites.

The largest contribution that photosite size makes to sensor SNR is that smaller photosites tend to produce less background noise.

The reason that high MP sensors tend to have worse high ISO noise performance, (but often not worse low ISO noise performance) has to do with the number (not the size) of photosites. Reading more photosites leads to more read noise.

Click to expand...

Click to expand...

ARClark said:

That complaint is grossly overstated. It may not perform quite as well as unprocessed or lightly processed D5 or D6 bodies for reasons others have stated, although when down sampled to 21 mp it gets close.

As to the Z9 files not responding to Topaz Denoise, I offer these examples to the contrary. Both of these were shot in HE* Raw and processed by Topaz Denoise as a plugin to PS. Both were shot handheld at ISOs of 10,000 and 25,600, respectively. I photographed these in January 2022 shortly after getting the camera. I was testing both the Z9's ability to focus in low light (second image) and to see how it handled high ISOs. I was impressed on both counts.


Ruby-crowned kinglet with Z9 and 500mm PF at ISO 10,000


Ruby-crowned kinglet with Z9 and 500E FL, TC-20E iii at ISO 25,600

The second photo was very challenging as the camera with a 500E FL f/4 and 2x TC had to hunt in the low light. But I was fortunate to find a cooperative ruby-crowned kinglet that paused just long enough for the AF to catch up. This shot would have been all but impossible with my D850 with its marginal AF capabilities at f/8 maximum aperture and with an almost black OVF.

One note on the Topaz Denoise AI processing of high ISO HE* files (8-10K and higher), it does tend to produce blotchiness in OOF areas which required followup PP to remove.

Click to expand...

GuyDagar said:

ARClark said:

That complaint is grossly overstated. It may not perform quite as well as unprocessed or lightly processed D5 or D6 bodies for reasons others have stated, although when down sampled to 21 mp it gets close.

As to the Z9 files not responding to Topaz Denoise, I offer these examples to the contrary. Both of these were shot in HE* Raw and processed by Topaz Denoise as a plugin to PS. Both were shot handheld at ISOs of 10,000 and 25,600, respectively. I photographed these in January 2022 shortly after getting the camera. I was testing both the Z9's ability to focus in low light (second image) and to see how it handled high ISOs. I was impressed on both counts.


Ruby-crowned kinglet with Z9 and 500mm PF at ISO 10,000


Ruby-crowned kinglet with Z9 and 500E FL, TC-20E iii at ISO 25,600

The second photo was very challenging as the camera with a 500E FL f/4 and 2x TC had to hunt in the low light. But I was fortunate to find a cooperative ruby-crowned kinglet that paused just long enough for the AF to catch up. This shot would have been all but impossible with my D850 with its marginal AF capabilities at f/8 maximum aperture and with an almost black OVF.

One note on the Topaz Denoise AI processing of high ISO HE* files (8-10K and higher), it does tend to produce blotchiness in OOF areas which required followup PP to remove.

Click to expand...

Nice shots, Alan!

I second the emotion. I was able to get some keepers of this Northern Flicker shooting through my window, which may explain why it's not as sharp as it should be. And likewise, it would have been impossible with my D850. I used Topaz Photo AI for the noise. ISO 22800. But the Z9 nailed the bird's eye.

Looks like the TC-20e III works well with your 500mm

Click to expand...

PerfectPoms said:

There are some photographers who are very happy with their superiority to us when they slam us for using the 'spray and pray' technique

Click to expand...

PerfectPoms said:

- but those of us who shoot at 20 fps have learned that out of those 20 photos, there will always be a couple of photos that are more in focus, and with better wing/leg positioning.

Click to expand...

Visionist said:

A friend of mine recently said her Z9 is terribly disappointing at ISOs above 4000 or so. She said it does not compare to her D6, and that Z9 pics at high ISOs are simply unusable, even using DxO PureRAW or Topaz Photo AI. She shoots BIF and wildlife typically in low light in the late afternoons and early morning. I've seen some of the RAW pics and was quite surprised at the amount of noise, especially some she took at ISO 25600. Can anyone corroborate her opinions? If it's true, I'll pass on the Z9 myself.

Click to expand...

Visionist said:

A friend of mine recently said her Z9 is terribly disappointing at ISOs above 4000 or so. She said it does not compare to her D6, and that Z9 pics at high ISOs are simply unusable, even using DxO PureRAW or Topaz Photo AI. She shoots BIF and wildlife typically in low light in the late afternoons and early morning. I've seen some of the RAW pics and was quite surprised at the amount of noise, especially some she took at ISO 25600. Can anyone corroborate her opinions? If it's true, I'll pass on the Z9 myself.

Click to expand...

Philip Eihuyar said:

chambeshi said:

The established knowledge is that photosite area correlates inversely with SNR is established on the differences in captured photons.

https://www.lensrentals.com/blog/2012/02/sensor-size-matters-part-2/

A published analysis comparing image sensor specifications supports the principle that "...Decreasing pixel size increases the likelihood that photon noise will become visible....".

Catrysse, Peter B., and Brian A. Wandell. "Roadmap for CMOS image sensors: Moore meets Planck and Sommerfeld." Digital Photography. Vol. 5678. SPIE, 2005.

It follows that the effects of higher SNR from larger photosites scales upwards to influence overall sensor noise as a function of pixel count. Compared to smaller photosites (eg 45mp sensor) the advantage of large photosites becomes evident in less sensor noise as ISO increases - eg in low light conditions. In contrast, smaller pixels generate proportionally more noise at higher ISOs. At higher ISO > 6400, the larger photosites of the D6 (basically = D5) 20mp sensor confer a distinct advantage, which Nikon optimized for lowlight photography.

https://www.imaging-resource.com/ne...-inside-nikons-super-secret-sensor-design-lab

The details of the D5 sensor architecture are propriety albeit the tech teardown with Xray data etc is available behind a paywall. But they presumably entail the optimization of microlenses and colour filters etc (as much as Etchels was permitted by Nikon to see for his IR article). There must be sound reasons why Nikon designed the D5 sensor on the foundation of 6.41 µm photosites and not on a D800 spec sensor (4.88 μm) or even smaller photosites. It follows the sensor engineers exploited the ability of these larger photosites to capture more light with higher SNR at higher ISOs, and confer a higher SNR on the lowlight advantage of the 20mp sensor.

The applications of these advantages in lowlight imaging extend beyond terrestrial photography into cutting edge scientific applications. It appears NASA scientists did their homework carefully in selecting the Nikon D5 (not the D800 sensor) for the ISS:

https://www.sciencedirect.com/science/article/pii/S003442572100331X

Notwithstanding that we can presume Nikon designed the Z9 BSI stacked sensor several years after that in the D5, the evidence is clear in the comparative sensor data (eg Photonstophotos) and Z9 images at higher ISOs that optimization of sensor micro architecture faces finite limits to mitigate the constraints of smaller photosite on noise.

Click to expand...

How great a difference in photon/light capture is there between a full frame sensor of equal pixel count with BSI tech design and non-BSI sensor? 30% more photon capture? 50%? This is where I feel the direct comparison between the D5 D6 and Z9 needs to include these different tech developments in the actual sensors of cameras. In addition, does the extra electronics in the 'stacked' sensor contribute to 'read noise'? To me, it is complicating my understanding because I think BSI is a big deal at the end of the process in image quality/noise.

Click to expand...

Philip Eihuyar said:

Philip Eihuyar said:

FingerPainter said:

No. For it to be "knowledge", it has to be true, and it isn't. It has been a common mistaken belief of many misinformed people for many years now. Photosite size has very little to do with the SNR performance of whole sensors. it has nothing to do with the shot noise of a sensor. A sensor captures the same number of photons at a given exposure regardless of how many photosites it is divided into.

People who mistakenly believe that larger photosites lead to less noisy images mostly don't understand that the noisiness of the image is not the same as the noisiness of an individual pixel. They mistakenly think that since a larger photosite captures more light, and therefore has a higher SNR, that the whole image must therefore have a higher SNR. They forget to account for the fact that there are more of the smaller photosites on a sensor of the same size, and this extra number of photosites exactly makes up for the smaller size of the photosites. A sensor of a given size captures the same number of photons at a given exposure regardless of the size of its individual photosites.

The largest contribution that photosite size makes to sensor SNR is that smaller photosites tend to produce less background noise.

The reason that high MP sensors tend to have worse high ISO noise performance, (but often not worse low ISO noise performance) has to do with the number (not the size) of photosites. Reading more photosites leads to more read noise.

Click to expand...

Click to expand...

Click to expand...

chambeshi said:

Philip Eihuyar said:

chambeshi said:

https://www.lensrentals.com/blog/2012/02/sensor-size-matters-part-2/

A published analysis comparing image sensor specifications

https://www.sciencedirect.com/science/article/pii/S003442572100331X

Notwithstanding that we can presume Nikon designed the Z9 BSI stacked sensor several years after that in the D5, the evidence is clear in the comparative sensor data (eg Photonstophotos) and Z9 images at higher ISOs that optimization of sensor micro architecture faces finite limits to mitigate the constraints of smaller photosite on noise.

Click to expand...

How great a difference in photon/light capture is there between a full frame sensor of equal pixel count with BSI tech design and non-BSI sensor? 30% more photon capture? 50%? This is where I feel the direct comparison between the D5 D6 and Z9 needs to include these different tech developments in the actual sensors of cameras. In addition, does the extra electronics in the 'stacked' sensor contribute to 'read noise'? To me, it is complicating my understanding because I think BSI is a big deal at the end of the process in image quality/noise.

Click to expand...

Good questions. I don't know specifics, however the BSI architecture must improve efficiency as well as facilitating layered circuitry. Perhaps BSI architecture is more relevant in the design of higher resolution sensors, which would arguably help to optimize the layout and photon capture of the smaller photosites in a 45mp sensor, compared to a 20mp sensor?

The section entitled "What exactly do Nikon’s sensor designers do?" (In the article above in Imaging Resource) appears to be a fairly accurate summary of the role of Nikon's sensor laboratory in optimizing respective image sensors.

One of the insights revealed in the IR article on Nikon's sensor engineering is how the engineer's applied research pays off in optimizing well performance - to maximize light capture. In the case of the D5 sensor, the engineers optimized high sensitivity and quantum efficiency.

In other words they built on the ability of the larger photosites in the 20mp sensor to maximize the capture and efficient quantum conversion of the minimum of photons. This is also likely where Nikon's “gapless” microlens technology plays a key role, and the proof is evident in that this 20mp sensors still rank so high in the industry. The evidence is plain to see in low light sensitivity of the D5 and D6 cameras.

Click to expand...

wildlifr said:

Click to expand...

PerfectPoms said:

I was in no way accusing you of 'spray and pray' - but I was just pointing out that the people who accuse others of this technique are missing out on the benefits of more photos to look through later to pick out the best one.

Click to expand...

PerfectPoms said:

Another option would be to try stacking the 1.4 and 2 x TC on the Sigma lens and see how well that works on mirrorless.

Click to expand...

Visionist said:

A friend of mine recently said her Z9 is terribly disappointing at ISOs above 4000 or so. She said it does not compare to her D6, and that Z9 pics at high ISOs are simply unusable, even using DxO PureRAW or Topaz Photo AI. She shoots BIF and wildlife typically in low light in the late afternoons and early morning. I've seen some of the RAW pics and was quite surprised at the amount of noise, especially some she took at ISO 25600. Can anyone corroborate her opinions? If it's true, I'll pass on the Z9 myself.

Click to expand...

GuyDagar said:

Visionist said:

A friend of mine recently said her Z9 is terribly disappointing at ISOs above 4000 or so. She said it does not compare to her D6, and that Z9 pics at high ISOs are simply unusable, even using DxO PureRAW or Topaz Photo AI. She shoots BIF and wildlife typically in low light in the late afternoons and early morning. I've seen some of the RAW pics and was quite surprised at the amount of noise, especially some she took at ISO 25600. Can anyone corroborate her opinions? If it's true, I'll pass on the Z9 myself.

Click to expand...

Your friend must be a novice. I hope she reads all these replies.

My final take:

I am new to the Z9 and a D6 user, as well as a Z7, Z6 II, and D850 user. I am loving the Z9. Is it perfect? No. Not that I can tell, yet, but I haven't had a chance to really put it through its many paces. No digital camera I've used in 20 years, going back to the D100 thru the first Z bodies, have done it all, have been perfect. Each has had its weaknesses, each has had its strengths. As for DSLRs, the D850 may be closest to doing everything really well, but it falls short too. I'm still of the opinion that the D800E takes the sharpest photos at base ISO, a little sharper and better than the D810, and that may have something to do with the way Nikon negated the effects of the low-pass filter rather than eliminating it. In any case, every Nikon, especially from the D3s on up has been capable of producing images that match or exceed the best photos taken with a Nikon F2 or F3 35mm and Kodak Kodachrome 64. They are simply amazing.

So in the end, as many have said, it really comes down to what you're shooting: shooting JPEGs for news and sports, shooting landscapes for large prints, shooting portraits and weddings for online albums or small prints, shooting birds in flight and wildlife—they all have special needs. Can the Z9 do it all? Do you know?

A couple recent photos I took, one with the Z9 and a similar one with the Z6 II make it a hard question to answer. And I think all this hubbub and criticism of one format versus another is kind of senseless. It's really the final image that matters, not how you got there, not which camera or lens you used, not the ISO, the PS plugins or anything else. It's only the final image that matters.

Like I said, I'm loving the Z9, but I will keep the D6, the Z6 II and the D850. They all do slightly different things, each better than the others.


The Z9's eye-recognition is amazing, it sticks like glue.


The Z6 II in low light, pretty hard to beat!

Note to critics: both images were processed similarly in Lightroom, cropped as necessary, processed with Topaz Photo AI, then edited in Photoshop and using a PS action I created, both reduced to the same final size for the web: 2160x1440 72 dpi. Another action added rounded corners and the specs text at bottom. Which is better? I don't know.

And, finally, I'll say this: I love photography and I love the gear, all of it. If I could afford it, I would have a copy of every great camera Nikon ever made, and maybe a Canon or two as well. Yeah, maybe a Hasselblad and a Pentax too.


Nikon D100

Click to expand...

Alan Clark Nature Photography

I'm a retired environmental professional turned nature photographer with a special interest in bird photography. Welcome to my site.

arclark.smugmug.com