Phase One IQ5 – Sony 247 MPX – 3:2 – will you get it?

[Paul Spinnler]

Alright, I think we are partly talking past each other. Also thanks for relaying the ARRI paper. Some points:

I am not sure about DR "has nothing to do with highlights but only with a noise limit in shadows." In my view this is inaccurate because dynamic range fundamentally refers to the ratio between the sensor's maximum capacity (full-well capacity) and its minimum detectble signal (noise floor). This inherently includes both highlights and shadows.
The ARRI white paper explicitly defines dynamic range as the ratio between signal saturation (highlights) and sensitivity threshold (shadows) on page 8 in the example. Therefore, it is incorrect to ignore the highlight aspect, as it is equally crucial in defining the total DR of a sensor.

Regarding cinema vs. photo: It’s important to clarify that Log processing is a post-capture tne mapping, not something that fundamentally changes the inherent sensor-captured DR.
Log processing is used to compress the range of highlights and shadows into a smaller bit depth for post cinema workflows and data storage given you record a huge amount of data, ensuring that the recorded image retains as much of the original dynamic range as possible when stored in lower bit-depth formats (such as 10-bit or 12-bit). It does not alter the sensor's ability to capture DR -it only maks the captured DR easier to manipulate in post-production.

Still photography in the case of an IQ4 relies on 16-bit RAW files that capture all linear sensor data without compression. IIQ files are 16-bit linear RAW files, meaning that they inherently store all the dynamic range captured by the sensor without the need for Log compression or encoding and Log processing does not enhance DR, it just stores it efficiently.

On the specific architecture, do you have a link on the S3 architecture in specific? I wanted to research this at one point, but only found generalized mention of dual gain architecture, so would be eager to learn more about the specific implmentation.

On BSI / stacked: my terminology was not correct - what I meant was the back-side illumination architecture allows for more light gathering as there's a back illumination layer below. Stacked is indeed a term more reserved for real stacking as for example in some more recent CPU designs.

So I think on a linear basis we can compare sensors, excluding the file encoding approaches which are different between mediums because you DO NEED to compress minutes of 24 or more FPS vs. still shots where you can just drop all recorded scene info into a large 16bit information domain. Log encoding is essentially just a software trick to keep files manageable, when going down from 16 bit to a lower encoding for a continuous video format.

For cinema people its their main entry point to the file data as they need to then ofc work with video data in post - grading, cutting - and here you can only manage at 4k+ resolutions some ideally log encoded base files (instead of pure raw video data which would quickly fill up any HDD).

It remains to be seen what the specs of the new IQ5 are on the DR front.

And again - curious on any info you have re S3's implementation of dual gain.

 

[SrMphoto]

I am not sure about DR "has nothing to do with highlights but only with a noise limit in shadows."

To clarify:
When comparing two cameras, the DR number tells you something about the (deep) shadows. It tells you nothing about the highlights.

On the specific architecture, do you have a link on the S3 architecture in specific?

Manufacturers rarely write about the details of dual gain implementation. However, dual conversion gain is the established dual gain implementation for still photography as it gives the best results. A few hybrid still/video cameras (GH6 and GH7) implement dual output gain instead, and the still photography IQ suffers in those cameras.

 

[Paul Spinnler]

To clarify:
When comparing two cameras, the DR number tells you something about the (deep) shadows. It tells you nothing about the highlights.

Manufacturers rarely write about the details of dual gain implementation. However, dual conversion gain is the established dual gain implementation for still photography as it gives the best results. A few hybrid still/video cameras (GH6 and GH7) implement dual output gain instead, and the still photography IQ suffers in those cameras.

Thats not correct - dynamic range, by definition, encompasses both the shadows and the highlights. As per the ARRI white paper you referenced, dynamic range is explicitly defined as the ratio between signal saturation (the maximum signal level before clipping in the highlights) and the sensitivity threshold (the minimum signal level discernible above the noise floor in the shadows). Both can vary sensor by sensor based on the design architecture. Its on p7.

And what about the distinction between photo and cinema sensors you introduced - on a linear basis sensors are comparable ... no matter if they are put into a cinema camera body and their analog output is saved via conversion into a log encoded video file or in the case of a P1 back into a 16 bit raw still file. One can still measure technically the full well capacity and read noise. As a result one can express dynamic range in DB and it can vary between sensors both based on their starting and end points. Log encoding for post processing is just tone mapping of data into smaller bit formats for efficient post production and storage - has no impact on linear analog to digital output of sensors.

On the S3 from Leica - so you have no specific architecture details about the sensor actually? It sounded earlier like you knew what and how it is designed to extend DR, citing specific technology implementations.

 

[SrMphoto]

Thats not correct - dynamic range, by definition, encompasses both the shadows and the highlights. As per the ARRI white paper you referenced, dynamic range is explicitly defined as the ratio between signal saturation (the maximum signal level before clipping in the highlights) and the sensitivity threshold (the minimum signal level discernible above the noise floor in the shadows). Both can vary sensor by sensor based on the design architecture. Its on p7.

Can you tell me what it means for highlights if a camera A has 14 stops of DR and camera B has 12 stops? Nothing, it has no effect on the highlights.

On the S3 from Leica - so you have no specific architecture details about the sensor actually? It sounded earlier like you knew what and how it is designed to extend DR, citing specific technology implementations.

I could not find anything. There is no information that S3 uses dual output gain as you claimed either. We know that dual conversion gain (DCG) is a standard way to implement dual gain for still photography, that DCG gives better results for still, and that the S3’s measurement curve looks like a DCG (GH6 and GH7 have different looking P2P curves). It is safe to deduce that S3 uses DCG. It is unfounded to claim that S3 uses dual output gain.

 

[Paul Spinnler]

Can you tell me what it means for highlights if a camera A has 14 stops of DR and camera B has 12 stops? Nothing, it has no effect on the highlights.

Actually, it means a a lot for highlights, especially when it comes to handling complex lighting situations like an indoor scene with bright outdoor sunlight streaming in or a beach scene with reflections on the sea or from windows or metallic objects. I encounter limits with the IQ4 on sunny days often, actually, meaning ETTR, which is not a problem.

One example interior scene could be if sunlight comes through a window at about 80,000 lux, which is very bright, while other parts of the room receive diffused light at 10,000 lux. There could also be drk areas in the corners or under furniture at 10 lux. The contrast ratio between the bright sunlight and the deepest shadow in such a scene is 80,000:10, which equates to roughly 13 stops of dynamic range.

A camera with 12 stops DR will struggl in this environment. The operator might have to choose between preserving the highlight details from the sunlight or maintaining the shadow details - you cannot handle both effectively if highlights clip. As a result, the brightest parts, such as the sunlight streaming through the window or reflected off surfaces, will likely clip if you exposre for the shadows, turning into pure white with no detail. This is a key point where dynamic range is not just about shadows alone; it directly affects how well you can preserve bright highlights if you operate beyond the sensor's DR in a scene.

That's why we learned typically to expose to the right to preserve highlights as it is more obvious and bad looking to have blown out windows, sky or reflections, but then in very dark areas you just get mush, even with an IQ4. There's a reason why there's exposure bracketing even with digital backs for indoor architectural photography or landscape photography.

2 stops of additional DR mean the camera can discern finer gradations in the brightest parts of the image before clipping occurs. In real terms, this allows for subtle detail to be preserved in the highlights, like the variations in brightness from direct sunlight glinting off surfaces or texture in bright clouds outside the window.

The ARRI Alexa 35 was introduced with 17 stops of dynamic range precisely to handle these kinds of challenging, mixed-lighting scenarios. In cinematography, creating a visually pleasing highlight roll-off -where bright areas don't abruptly clip but instead transition smoothly to an off-white - is crucial for a natural, filmic look. With 17 stops, the Alexa 35 achieves a contrast ratio of 131,072:1, providing more than enough headroom to handle both the extreme highlights of bright outdoor sunlight and the deep shadows of an indoor environment. Just look at some of the intro videos, there's precisely some indoor scenes with outside light they show off.

I often shoot Portra 160 and I can see a clear difference in the rendition of the sky between the digital back and the film and I wish the back could handle a bit more highlights while correctly exposing the darks as well. Also DR will help with LCC of UWAs where you need to compensate for 2 or more stops of difference between the corners and centre, like in the case of the 28XL from SK.

Cameras like the Alexa 35 are designed for smooth highlight roll-off, where the transition from highlights to midtones is gradual, preserving the aesthetic look and preventing harsh clipping that looks unnatural.

The differece between 14 to 12 stops is very real.

On the S3 – Leica back in the day mentioned that the sensor has a dual gain structure – whether it uses gain switching or dual readout is unclear, but I agree it is more likely that it uses a sort of switching system rather then two readout circuits given the complexity of the latter, but we cannot know what has been implemented and how and it sounded to me back then that they did something proprietary as it is a custom chip.

 

[hasselblad 503cw]

I have not much acknowledge about what kind of display or media can demonstrate us the 17 stop or 14 stop DR image.
Does 17 stop or 14 stop image shall be translated or edited by some kind of curve or log math something?

 

[SrMphoto]

I have not much acknowledge about what kind of display or media can demonstrate us the 17 stop or 14 stop DR image.
Does 17 stop or 14 stop image shall be translated or edited by some kind of curve or log math something?

Newer display technologies (HDR) offer more than 12 stops of dynamic range.
However, the dynamic range is more about the noise in the shadows, i.e., how much noise will you see when lifting deep shadows in high-contrast scenes?
Depending on the scene, bracketing may still be necessary. This means the cameras do not provide sufficient dynamic range, which must be increased with bracketing (or frame averaging).

 

[SrMphoto]

Actually, it means a a lot for highlights, especially when it comes to handling complex lighting situations like an indoor scene with bright outdoor sunlight streaming in or a beach scene with reflections on the sea or from windows or metallic objects. I encounter limits with the IQ4 on sunny days often, actually, meaning ETTR, which is not a problem.

In high- and low-contrast scenes, you maximize the exposure so that the sensor saturates but does not clip relevant highlights (ETTR). In high-contrast scenes, you are burdened with deep shadows that can show a lot of noise once they are lifted. How much noise you see depends on the camera's DR. Highlights are not affected by the camera's DR; shadows are.

One example interior scene could be if sunlight comes through a window at about 80,000 lux, which is very bright, while other parts of the room receive diffused light at 10,000 lux. There could also be drk areas in the corners or under furniture at 10 lux. The contrast ratio between the bright sunlight and the deepest shadow in such a scene is 80,000:10, which equates to roughly 13 stops of dynamic range.

A camera with 12 stops DR will struggl in this environment.

Digital sensors do not clip shadows like negatives do. A 12-stop DR camera will show more noise in the shadows than a 14-stop DR camera, but highlights are not affected.

The operator might have to choose between preserving the highlight details from the sunlight or maintaining the shadow details - you cannot handle both effectively if highlights clip. As a result, the brightest parts, such as the sunlight streaming through the window or reflected off surfaces, will likely clip if you exposre for the shadows, turning into pure white with no detail. This is a key point where dynamic range is not just about shadows alone; it directly affects how well you can preserve bright highlights if you operate beyond the sensor's DR in a scene.

In proper exposure, clipping relevant highlights is not an option. You do not decide to clip if you have less DR. Instead, you can use bracketing, frame averaging, or noise reduction to handle the increased noise in the shadows.

That's why we learned typically to expose to the right to preserve highlights ...

ETTR is not about preserving highlights but maximizing the exposure and thus reducing the visible noise.

... as it is more obvious and bad looking to have blown out windows, sky or reflections, but then in very dark areas you just get mush, even with an IQ4. There's a reason why there's exposure bracketing even with digital backs for indoor architectural photography or landscape photography.

2 stops of additional DR mean the camera can discern finer gradations in the brightest parts of the image before clipping occurs. I

Assuming you expose with ETTR, two stops of additional DR have no influence on gradations in the highlights but on the detail and gradation in the shadows.

n real terms, this allows for subtle detail to be preserved in the highlights, like the variations in brightness from direct sunlight glinting off surfaces or texture in bright clouds outside the window.

The ARRI Alexa 35 was introduced with 17 stops of dynamic range precisely to handle these kinds of challenging, mixed-lighting scenarios. In cinematography, creating a visually pleasing highlight roll-off -where bright areas don't abruptly clip but instead transition smoothly to an off-white - is crucial for a natural, filmic look. With 17 stops, the Alexa 35 achieves a contrast ratio of 131,072:1, providing more than enough headroom to handle both the extreme highlights of bright outdoor sunlight and the deep shadows of an indoor environment. Just look at some of the intro videos, there's precisely some indoor scenes with outside light they show off.

The raw data stored in the raw file is linear. The post-processor handles any highlight roll-off and similar issues by applying appropriate curves to the demosaiced data.

I often shoot Portra 160 and I can see a clear difference in the rendition of the sky between the digital back and the film

I wish digital sensors could work like negative film: darks clip, but highlights can never clip. That makes the skies much nicer, IMO.

and I wish the back could handle a bit more highlights while correctly exposing the darks as well.

What does it mean to expose the darks correctly with digital sensors? You cannot clip relevant highlights, and you get the darks as good as the camera can deliver them (DR).

Also DR will help with LCC of UWAs where you need to compensate for 2 or more stops of difference between the corners and centre, like in the case of the 28XL from SK.

Cameras like the Alexa 35 are designed for smooth highlight roll-off, where the transition from highlights to midtones is gradual, preserving the aesthetic look and preventing harsh clipping that looks unnatural.

The differece between 14 to 12 stops is very real.

Yes, but it affects mainly darks. One can simulate the effect of 14 vs 12 stops of DR by taking a shot with the ETTR exposure setting and one with two stops below that ETTR exposure setting.

On the S3 – Leica back in the day mentioned that the sensor has a dual gain structure – whether it uses gain switching or dual readout is unclear, but I agree it is more likely that it uses a sort of switching system rather then two readout circuits given the complexity of the latter, but we cannot know what has been implemented and how and it sounded to me back then that they did something proprietary as it is a custom chip.

Based on the P2P measurement, it should be pretty clear that S3 is using DCG. FWIW, S3 uses the same sensor technology as M10-R.

 

[GrahamWelland]

3:2 would be a deal breaker for me.

 

[Hel]

Thank you for (hopefully)ending that off-topic DR discussion

 

[Paul Spinnler]

Thank you for (hopefully)ending that off-topic DR discussion

So 3:2 also killer for you?

Unfortunately, the 811 chip is the only one available

 

[BigBoy]

Will there be a new XF?

 

[Paul Spinnler]

Will there be a new XF?

100% not.

You'll get:

1) IQ5 based on Sony 247
2) Hardware update around the sensor with 2024 tech (better I/O, etc.)
3) Maybe one more XC lens, e.g. 70mm which is almost the same as Leica's Q3 43
4) XT XL so the SB adapter introduced for the 90 SW can be useful for stitching
5) Maybe more own lenses, designed by SK -> could be SK 72 MK II which is used in repro already and or other lenses like the 120 ASPH reloaded, etc.
6) If they are smart and we are lucky and EVF solution

Although 3:2 is a shocker, people might adapt over time, especially considering that a 54x40 aspect ratio crop in the 3:2 still is a 220 MPX file.

This means you are more working within the sweet spot of lenses and the 40 is the new 50, etc.

And in the end - its quite exciting a new tech back. So I assume the IQ5 will revive the category and push sales of tech cams and further increase SK glass prices.

The new IQ5 is also the ultimate one-shot film scanner, especially 35mm rolls. One shot 250 megapixel scans is hard to beat.

 

[anyone]

Now someone needs to show that 35mm image that contains data for such a scan Seriously, 100mp (or 60mp crops) are more than enough for 35mm scans.

It’s a different story for panoramic formats.

 

[BigBoy]

100% not.

You'll get:

1) IQ5 based on Sony 247
2) Hardware update around the sensor with 2024 tech (better I/O, etc.)
3) Maybe one more XC lens, e.g. 70mm which is almost the same as Leica's Q3 43
4) XT XL so the SB adapter introduced for the 90 SW can be useful for stitching
5) Maybe more own lenses, designed by SK -> could be SK 72 MK II which is used in repro already and or other lenses like the 120 ASPH reloaded, etc.
6) If they are smart and we are lucky and EVF solution

Although 3:2 is a shocker, people might adapt over time, especially considering that a 54x40 aspect ratio crop in the 3:2 still is a 220 MPX file.

This means you are more working within the sweet spot of lenses and the 40 is the new 50, etc.

And in the end - its quite exciting a new tech back. So I assume the IQ5 will revive the category and push sales of tech cams and further increase SK glass prices.

The new IQ5 is also the ultimate one-shot film scanner, especially 35mm rolls. One shot 250 megapixel scans is hard to beat.

Ok makes sense. So I assume then Hassy/Fuji/Leica will be releasing their new sensors as well next year then too?

 

[Paul Spinnler]

I would not bet on a quick release of the next gen sensor for the other platforms.

The 811 MX is cutting edge and as of now the only announced chip from that generation is the 247 chip. Historically Phase One had a tiny timing advantage so they were first and then Fuji came along with 100 megapixels, then the rest.

It will all depend on the release schedule and pricing from Sony. For example, as of now, no new crop MF chip in that pixel pitch has been announced. This means once announced, there's still time needed for implementation and on top pricing needs to fir the Fuji / Hasselblad requirements.

Sony right now can ask let's say 5k for the sensor from Phase One while a hihgly priced crop variant might not fit into the pricing system of current Fuji and Hassy cameras. If the chip costs 3k, the Hassy costs 8k and then there's margin in between and production cost - I am just saying that it could take a bit before we see this new chip come out in crop.

I suppose Sony also needs to ramp up the production of this one to see how yields are etc.

My best bet would be: early 2025 IQ5, then S4 with either 100 or 180 megapixels and then by end of year or beginning of the year after the other systems.

But we'll know once Sony releases new product spec sheets - right now there's only the 247 chip, meaning the other crop sensors are a bit off, still.

Only reason I think S4 could be directly 180 is because they are a huge customer of Sony and on top the S4 can be timed perfectly with the availability of the new chip, but all bets are off who gets first dibs on the crop chips –> once announced and we are not there yet.

Right now only game in town seems to be Phase with 250 beg. of 2025.

 

[dougpeterson]

I don’t think I would describe Phase One’s historic timing advantage as “tiny” looking at the timing for the actual shipping-in-meaningful quantity timelines for the crop 50mp, 645 100mp, crop 100mp, and 645 150mp sensors.

I was at the Hong Kong launch of the IQ250 (first with 50mp cmos) back. Hasselblad caught wind of the pending launch and made a rushed “announcement of an announcement” presumably so they could claim to be the first. But the IQ250 shipped within weeks and the Hassy equivalent shipped like a year later*. Hassy was also significantly later with actual shipping 100mp 54x40, and never did ship a 150mp system.

*this is off the top of my head - it’s been a few years now and I could be a bit off in either direction in my recollection; knowing you you’ll help out by tracking down the exact dates by scouring the threads here on getdpi

Hasselblad has been doing some really neat mid-high end products recently. But they’ve ceded the high-high end to P1. Whatever flagship comes next and whenever it comes I would expect P1 to be shipping it for quite some time before the next equivalent. For what it’s worth I don’t think it’s as “around the corner” time wise as this thread makes it sound.

Of course I could be wrong; I’m biased, and predictions are hard - especially about the future!

 

[Doppler9000]

What about the 40 MP chip in the X-H2 and X-T5 Fujis?

This would be ~230 MP in a 54x40mm sensor.

 

[rdeloe]

*this is off the top of my head - it’s been a few years now and I could be a bit off in either direction in my recollection; knowing you you’ll help out by tracking down the exact dates by scouring the threads here on getdpi

https://imgs.xkcd.com/comics/duty_calls_2x.png

 

[Paul Spinnler]

I don’t think I would describe Phase One’s historic timing advantage as “tiny” looking at the timing for the actual shipping-in-meaningful quantity timelines for the crop 50mp, 645 100mp, crop 100mp, and 645 150mp sensors.

I was at the Hong Kong launch of the IQ250 (first with 50mp cmos) back. Hasselblad caught wind of the pending launch and made a rushed “announcement of an announcement” presumably so they could claim to be the first. But the IQ250 shipped within weeks and the Hassy equivalent shipped like a year later*. Hassy was also significantly later with actual shipping 100mp 54x40, and never did ship a 150mp system.

*this is off the top of my head - it’s been a few years now and I could be a bit off in either direction in my recollection; knowing you you’ll help out by tracking down the exact dates by scouring the threads here on getdpi

Hasselblad has been doing some really neat mid-high end products recently. But they’ve ceded the high-high end to P1. Whatever flagship comes next and whenever it comes I would expect P1 to be shipping it for quite some time before the next equivalent. For what it’s worth I don’t think it’s as “around the corner” time wise as this thread makes it sound.

Of course I could be wrong; I’m biased, and predictions are hard - especially about the future!

Well I remember sth around end of 2018 for the IQ4 150 and sometime 2019 for the GFX100. I am not sure how quickly Fuji shipped the 100 megapixel systems and ofc we've seen with Hassy that it took years for them to come to 100 MPX.

So maybe a fair assumption is 2025 for P1 and the rest from 2026 onwards, which would put an S4 in an awkward launch cycle at the end of life of the prior gen sensor series, so curious how this will pan out.

You meant the crop chips not being around the corner, right?

And as said, there are no new chips on the Sony website, so the rest could come even later ...

Looks like to me that cutting edge peeps will have to wait quite a while for Fuji / Hassy to put out a "cheap" IQ5 - in the end, Sony will happily sell 250 chips for a while at ultra high margins for them ...