What am I missing? Comparing sensors of a9i to a9iii and a1

[AceDynamo]

BACKGROUND: I have the original a9 and shoot mostly high school basketball, volleyball, and football, which all typically have low light situations. Most of my photos are shot with f2.8 and at ISOs between 3200 and 10000.

CURRENTLY: I am interested in upgrading my camera body to either the a9iii or a1ii for the improvements in auto focus, burst rate, pre-capture, and some other misc improvements; HOWEVER, the analysis at Photons to Photos is giving me second thoughts.

SENSOR COMPARISON: If I am interpretting the charts below correctly, it appears the original a9 sensor is performing better than the a9iii and the a1 in the ISO range I typically shoot (ISO 3200 to 10000). See images of the charts I included below.

QUESTIONS: For those who have shot with the original a9 and either the a9iii or a1, how does the image quality compare? What am I missing in interpretting these charts? Is the a9 image quality really better?

BOTTOM LINE: I am trying to weigh any tradeoffs if I upgrade.

Thanks for any insight!

Original A9 better DR at ISOs from 3200 to 12800?

Original A9 better lower noise at ISOs from 3200 to 12800?

 

[jonas ar]

NAwlins Contrarian wrote;

There is no way to tell whether baked-in noise reduction affected an image because you don't have an otherwise-identical version without that noise reduction to compare. Unless you have an otherwise-identical camera that does not bake in noise reduction, you're at best guessing

this is not true. See for example here: https://blog.kasson.com/a7riii/the-sony-a7riii-eats-stars/

And here https://www.photonstophotos.net/Gen...2D_Fourier_Transforms_for_Sensor_Analysis.htm

 

[NAwlins Contrarian]

Baked in NR or fudged ISO don't matter if they don't affect the image.

There is no way to tell whether baked-in noise reduction affected an image because you don't have an otherwise-identical version without that noise reduction to compare. Unless you have an otherwise-identical camera that does not bake in noise reduction, you're at best guessing

this is not true. See for example here: https://blog.kasson.com/a7riii/the-sony-a7riii-eats-stars/

And here https://www.photonstophotos.net/Gen...2D_Fourier_Transforms_for_Sensor_Analysis.htm

Let my restate / modify my prior statement, because you've raised a fair point: there is no way to conclude that "Baked in NR ... do[es]n't matter" because it didn't "affect the image," which was the claim to which I had responded. One can't prove or even really evaluate the negative claim. The baked-in noise reduction may visibly affect the image, and often there is no way to tell, and certainly no way to directly compare.

Conversely, you are correct that sometimes we can observe things that baked-in noise reduction appears to have caused. The 'star eater' issue is an example.

 

[TN Args]

I also think that the OP might be better served looking at DPR test charts at the various ISO values and comparing them for noise, than the Photons to Photos charts.

...for example, at ISO 25600 on the DPR test chart raw files.

top row: A9 III, A9

second row: A1, R5 II

These are out of FastRawViewer, set to no sharpening or NR

cheers

Do me a favour, check the shutter speeds of those raw files in the comparison and tell me what you find…

Sure.

A9III 1/10,000

The others 1/5,000

All at ƒ/5.6 (although three different lenses used might have different T-stops)

See note at bottom which explains the variation.

IMO the issue is not so much T-stops of the lenses used, especially at moderate apertures; the issues are that (1) some manufacturers really fudge the ISO ratings and (2) some manufacturers bake noise reduction into the raw files.

We could do things like look at what it takes to get the same degree of ETTR for an exposure, or shoot various cameras at the same nominal exposure under the same lighting and compare the average lightness of a middle-gray patch.

For noise what matters is the exposure. ISO is not part of exposure and hence mostly irrelevant. Why dpreview used a faster exposure time for a9iii is a bit mysterious and invalidates the result unless the exposure was adjusted by raising scene brightness or using a larger aperture?

Something is amiss. The DPR widget says all 4 images were at 1/80s and ƒ/5.6. I will look more closely into it when I get back to the PC I used for downloading and viewing.

[edit: OK the fast shutter speeds are for the daylight simulation lighting test scene images and the slow ones are for the low light test image. Maybe I will repeat the comparison with the low light test images tomorrow. At least then they will all be same exposure. The low light version is probably more representative anyway...]

cheers

OK I have looked more closely and nothing is amiss after all. The above comparison images that I posted are 'valid'.

DPR are adjusting the scene brightness at high ISO test scenes, and in the end all cameras are getting the same exposure to light. DPR mention this in their article on the test scenes: "At higher ISOs, we reduce the illumination of the scene by up to two stops if a camera doesn't offer sufficiently fast shutter speeds to allow correct exposure. If this still isn't sufficient, we then stop down the camera's aperture, again ensuring that the net effect of illumination, shutter speed and aperture values are consistent across cameras."

Although they mention the top shutter speed of each camera, I suppose there could be other reasons too. The main point being that they are being careful and getting it right. Whew!

Makes sense. My personal tests were done a bit differently, as I used the same lens, the same lighting, and all the exact same camera settings between the cameras, then I used Lightroom to compare iso 6400 & 12,800:

I had a look at your video; great job you did there. A couple of points came to mind.

The 1/16,000s limit with your ƒ/1.2 lens is a bit of a bummer. For that lens, an ND or polarising filter seems the best option unless the firmware is updated.

The A9 III's high ISO IQ matching the A7C is great to see. But I do have a suggestion for when you resize files for comparing A9 III with A1. Resize upwards, not downwards, and resize both cameras' files. For example, maybe to 60 MP. The act of resizing has consequences on IQ, so resizing only one file is advantageous for the camera that is left at its native MP.

cheers

 

[f1point4andbethere]

I also think that the OP might be better served looking at DPR test charts at the various ISO values and comparing them for noise, than the Photons to Photos charts.

...for example, at ISO 25600 on the DPR test chart raw files.

top row: A9 III, A9

second row: A1, R5 II

These are out of FastRawViewer, set to no sharpening or NR

cheers

Do me a favour, check the shutter speeds of those raw files in the comparison and tell me what you find…

Sure.

A9III 1/10,000

The others 1/5,000

All at ƒ/5.6 (although three different lenses used might have different T-stops)

See note at bottom which explains the variation.

IMO the issue is not so much T-stops of the lenses used, especially at moderate apertures; the issues are that (1) some manufacturers really fudge the ISO ratings and (2) some manufacturers bake noise reduction into the raw files.

We could do things like look at what it takes to get the same degree of ETTR for an exposure, or shoot various cameras at the same nominal exposure under the same lighting and compare the average lightness of a middle-gray patch.

For noise what matters is the exposure. ISO is not part of exposure and hence mostly irrelevant. Why dpreview used a faster exposure time for a9iii is a bit mysterious and invalidates the result unless the exposure was adjusted by raising scene brightness or using a larger aperture?

Something is amiss. The DPR widget says all 4 images were at 1/80s and ƒ/5.6. I will look more closely into it when I get back to the PC I used for downloading and viewing.

[edit: OK the fast shutter speeds are for the daylight simulation lighting test scene images and the slow ones are for the low light test image. Maybe I will repeat the comparison with the low light test images tomorrow. At least then they will all be same exposure. The low light version is probably more representative anyway...]

cheers

OK I have looked more closely and nothing is amiss after all. The above comparison images that I posted are 'valid'.

DPR are adjusting the scene brightness at high ISO test scenes, and in the end all cameras are getting the same exposure to light. DPR mention this in their article on the test scenes: "At higher ISOs, we reduce the illumination of the scene by up to two stops if a camera doesn't offer sufficiently fast shutter speeds to allow correct exposure. If this still isn't sufficient, we then stop down the camera's aperture, again ensuring that the net effect of illumination, shutter speed and aperture values are consistent across cameras."

Although they mention the top shutter speed of each camera, I suppose there could be other reasons too. The main point being that they are being careful and getting it right. Whew!

Makes sense. My personal tests were done a bit differently, as I used the same lens, the same lighting, and all the exact same camera settings between the cameras, then I used Lightroom to compare iso 6400 & 12,800:

I had a look at your video; great job you did there. A couple of points came to mind.

The 1/16,000s limit with your ƒ/1.2 lens is a bit of a bummer. For that lens, an ND or polarising filter seems the best option unless the firmware is updated.

The A9 III's high ISO IQ matching the A7C is great to see. But I do have a suggestion for when you resize files for comparing A9 III with A1. Resize upwards, not downwards, and resize both cameras' files. For example, maybe to 60 MP. The act of resizing has consequences on IQ, so resizing only one file is advantageous for the camera that is left at its native MP.

cheers

Hey thanks, I appreciate the comment.

About your suggestion for resizing upwards; that’s unfortunately not possible. My resizing is done during the dng conversion process, which doesn’t allow up scaling.