Pentax 645Z vs A7RII IQ and bang for the buck

[Chris Giles]

I'm happy with it, need to play about some more but initial stuff looks great.

 

[Ken_R]

I'm happy with it, need to play about some more but initial stuff looks great.

View attachment 93520View attachment 93521View attachment 93522

from what camera is that?

 

[Chris Giles]

from what camera is that?

Canon 5DSr Ken.

 

[Kolor-Pikker]

from what camera is that?

This whole thread Chris has been talking about waiting for his 5DSR to arrive so that he can test it out. Edit: beaten

+100 shadows is nothing though, add +4 exposure as well to really test it out, although even on this web-size jpeg the shadows in the bushes in the background already look fairly crunchy. Canon cameras also have some DR in the highlights, so add some negative setting to the highlights.

 

[Ken_R]

This whole thread Chris has been talking about waiting for his 5DSR to arrive so that he can test it out. Edit: beaten

+100 shadows is nothing though, add +4 exposure as well to really test it out, although even on this web-size jpeg the shadows in the bushes in the background already look fairly crunchy. Canon cameras also have some DR in the highlights, so add some negative setting to the highlights.

Yep, +1 by itself is not much.

Canons start to break down at about +3 exposure and at +4 and beyond they get really ugly. Of course the 5DS/R have much more resolution/pixels so one can use noise reduction more aggressively and still retain a lot of detail. But still, color quality in the shadows goes down quickly. The stock Adobe profile for the 5DS/R looks fairly contrasty, I would dial down contrast quite a bit also.

 

[Chris Giles]

ISO800

First image SOOC
Second image +2 exposure and +100 shadows
Third image +4 exposure and +100 shadows

Best I have right now, will get something real ugly for you in a moment.

 

[Chris Giles]

First image, ISO100 -4 stops.
Second image +4 exposure in post
Third image +100 shadows
Fourth image, close up.

As with all my cameras I don't sharpen, use noise reduction or anything like that.

 

[Chris Giles]

Processed to my taste (but still +4 stops). We're in interesting times with cameras. I'm particularly happy with the colours of the Canon, more so than my Pentax and A7ii. It'll be interesting to see broader use of this with skin tones over the next few weddings.

 

[jerome_m]

I am not really sure I understand the point of underexposing by 4 EV only to push by the same value afterwards.

...although, I do that all the time with my antique Hasselblad MF. It does not have an analogue amplifier, so any ISO above base ISO (which is ISO 50) is digital push. ISO 50 + 4 EV is ISO 400, which I sometimes use. Looks reasonably fine to me. But my Hasselblad is a CCD camera, designed in 2008, 7 years ago.

 

[Chris Giles]

I am not really sure I understand the point of underexposing by 4 EV only to push by the same value afterwards.

...although, I do that all the time with my antique Hasselblad MF. It does not have an analogue amplifier, so any ISO above base ISO (which is ISO 50) is digital push. ISO 50 + 4 EV is ISO 400, which I sometimes use. Looks reasonably fine to me. But my Hasselblad is a CCD camera, designed in 2008, 7 years ago.

Quite, I'm the same with my 645z, expose for the highlights and then push the rest. I tend to expose correctly so don't need the push files so much.

 

[jerome_m]

Quite, I'm the same with my 645z, expose for the highlights and then push the rest. I tend to expose correctly so don't need the push files so much.

I am pretty sure you expose correctly! I understand you did that just for the test.

What I wanted to point out is that digital pushing by +4 EV was already done as a standard function in digital backs in 2008, since this is the way my antique camera operates when set to ISO 400. If Hasselblad built that as a standard function, I think that they thought it was already good enough at the time.

So, basically, your test shows that your camera can do what a digital back could already do in 2008.

(Note that I am discussing the test, not the particular camera)

 

[Chris Giles]

I am pretty sure you expose correctly! I understand you did that just for the test.

What I wanted to point out is that digital pushing by +4 EV was already done as a standard function in digital backs in 2008, since this is the way my antique camera operates when set to ISO 400. If Hasselblad built that as a standard function, I think that they thought it was already good enough at the time.

So, basically, your test shows that your camera can do what a digital back could already do in 2008.

(Note that I am discussing the test, not the particular camera)

I know

I thought with CCDs the signal beyond ISO100 was amped by software anyway and CMOS the amplification was carried out at the point of capture.

 

[jerome_m]

I thought with CCDs the signal beyond ISO100 was amped by software anyway and CMOS the amplification was carried out at the point of capture.

Not really. The necessity for analogue amps comes from using an ADC with insufficient bit depth.

CCD or CMOS of about 6 µm pitch have a full well of about 50K photons. If you use a 16-bits ADC, you can count from 0 to 65535, so you can count the whole dynamic range of your sensor without needing analogue adjustment.

If you only have a 12-bits ADC, you can count from 0 to 4095, or by analogue amping down by a factor of 2, from 2 to 8191, or from 4 to 16385, or from 16 to 32767, or from 32 to 65535 (full well).

Early DSLRs had only 12-bits ADCs, so they needed an analogue amplifier. Hasselblad chose to use a high-end 16-bits ADC, so did not really need an analogue amplifier. On CMOS sensors, the ADCs are directly on the chip and the constraints of integration mean that using really fancy ones is difficult. Early CMOS ADCs were also only 12-bits.

 

[Kolor-Pikker]

On CMOS sensors, the ADCs are directly on the chip and the constraints of integration mean that using really fancy ones is difficult.

Specifically, Sony sensors have on-chip A/D conversion, which is why they have so little read noise compared to other manufacturers. I don't know if there are other companies that have adopted this technology, but you need to have a sufficiently small manufacturing process to put circuitry directly on the sensor, and Canon refuses to budge from their tried and tested methods that are too large to put anything on-sensor except pixels. They recently improved this and now have on-sensor PDAF, but still no A/D conversion, meanwhile Sony is pushing even further by using copper wiring and BSI to maximize the speed at which photons are read off the sensor and converted to data.

This is a big shame, since if we were to calculate the dynamic range of the 1DX recursively based on it's linear noise performance from ISO800~12800, it should have had even more DR than the D810 at ISO100 by nearly a stop. As it stands, the 1DX loses just around 0.5 stops of DR going from ISO100 to ISO800, because the read noise level significantly exceeds sensor noise.

 

[Pradeep]

The A7rii is certainly an interesting camera with a high possibility to rip a decent share of the MF market but all the time the sensor is 135 it'll miss a certain look and feel to the image.
Saying that, ask me in two weeks when I've done side by sides of the 5Dsr and the 645z at a few weddings because if I can dump the 645z for the lighter setup and am able to go back to a single line of lenses I will. Just so long as the images don't lose something.

Chris, I sold my 5DSR two weeks ago, after having tested it against the 645Z (there is a thread on it here).

At the end of the day, the camera needs to be coupled with a lens for the final results and in my comparison, not only was the Pentax better at pulling up shadows (no surprise there), but the best Canon glass I have (the 16-35 f4 IS and the 24-70 f2.8 MkII) could not match the Pentax with its legacy glass (the FA 45-85 and FA75) in center AND particularly corner sharpness, even when stopped down.

So for me, cost not being the most important factor (anybody who has owned a Phase IQ180 will agree that that was one helluva poor bang for the buck), the final IQ battle was won by the Pentax. The weight/heft/ease of use etc are secondary issues.

FWIW, I've also owned the Canon 24 TSE MkII and was not blown away by its IQ either. So the argument that the A7RII will be super sharp and better than the Pentax because you can put the Canon TSE lenses on it does not hold water for me.

 

[Pradeep]

Yep, +1 by itself is not much.

Canons start to break down at about +3 exposure and at +4 and beyond they get really ugly. Of course the 5DS/R have much more resolution/pixels so one can use noise reduction more aggressively and still retain a lot of detail. But still, color quality in the shadows goes down quickly. The stock Adobe profile for the 5DS/R looks fairly contrasty, I would dial down contrast quite a bit also.

I tested my 5DSR against the Pentax 645Z going from underexposure to overexposure and found that the Canon, even at ISO 100 was unable to pull up the shadows as cleanly. For a fair comparison you will need to have the same scene/lighting set up for both cameras and then see how much the image can be manipulated and what the results are.

I have an A7RII on order so yes, I too am very interested in getting a package that is not cumbersome and will not break the back or the bank.

These are interesting times.....

 

[D Fuller]

Not really. The necessity for analogue amps comes from using an ADC with insufficient bit depth.

CCD or CMOS of about 6 µm pitch have a full well of about 50K photons. If you use a 16-bits ADC, you can count from 0 to 65535, so you can count the whole dynamic range of your sensor without needing analogue adjustment.

If you only have a 12-bits ADC, you can count from 0 to 4095, or by analogue amping down by a factor of 2, from 2 to 8191, or from 4 to 16385, or from 16 to 32767, or from 32 to 65535 (full well).

Early DSLRs had only 12-bits ADCs, so they needed an analogue amplifier. Hasselblad chose to use a high-end 16-bits ADC, so did not really need an analogue amplifier. On CMOS sensors, the ADCs are directly on the chip and the constraints of integration mean that using really fancy ones is difficult. Early CMOS ADCs were also only 12-bits.

This is a great post. It helps me understand what issues really contribute to and control noise in sensor design. It's really very similar to the relationship of analog s/n to bit depth in audio recording.

 

[ErikKaffehr]

Hi,

There are a couple of points missing.

One is that modern CMOS sensors seem to have increased in full well capacity. I don't have actual numbers but I would guess that say Nikon D810 and say Phase One IQ have similar FWC (number of electron charges stored).

The other point is that the maximum meaningful amount of information from a sensor is FWC / readout noise. Readout noise for a Phase One IQ 180 may be around 12 electron charges. So the amount of information is really around 1:3800, say 1:4000. To represent that range you need 12 bits. So the output from the ADC is 16 bit but 4 bits of that is noise.

With recent CMOS sensors Sony and other vendors moved the ADCs on the chip. Normally there is an ADC for each column, so a normal 24 MP sensor has 6000 ADCs. Those ADCs are probably simple ramp type devices and can have long conversion times as each ADC only handles 4000 pixels.

Noise levels of on sensor ADCs may go down to 2 electron charges. Using the same source of info as for the the IQ-180 we find that a Sony Alpha 99 has a full well capacity of 64682 and and a readout noise of 2 electron charges. That means the useful information is around 60000/2 = 30000
which corresponds to 13.8 bits, that is close to 14 bits. In real world the Sony A99 seem to deliver around 13 bits of information, though. Both Pentax and Nikon seems to make better use of those Sony sensor then Sony themselves.

Now, DxO mark actually measures this. It is pretty much the same as engineering DR in "screen mode", this figure is 11.89 EV (12 bits) for the IQ-180 and 13.15 for the Sony Alpha 99. So the IQ-180 is actually a 12 bit device while the Sony Alpha is essentially a 13 bit device.

The image below was shot on the Sony Alpha 99

And this one was shot on my P45+

Both of these images were shot so they would retain some highlight detail in the windows.

Now, lets look at the Piano at the bottom of each image, first the Sony Alpha 99:

And the P45+:

These are not test shots under controlled circumstances but real world exposures based on ETTR using in camera histograms. The exposure on the Sony may have been more optimal. I tried to process both images similarly. Note that the P45+ image is 39 MP while the Sony Alpha 99 image is 24MP, downscaling the P45+ image to 24MP would reduce noise to some extent. But this is more like intended to illustrate dynamic range and bits needed rather than comparing different sizes of sensors.

The raw images are here, these are DNGs with original raw image embedded, can be extracted using Adobe DNG converter.
http://echophoto.dnsalias.net/ekr/Articles/DRArticle/Lockenhous/20140617-CF045290.dng
http://echophoto.dnsalias.net/ekr/Articles/DRArticle/Lockenhous/20140617_lumariver.dng

Just to say, I have not seen spec sheets for the IQ-180 sensor, but the Dalsa Specsheet for IQ-160 sensor gives around 12 bits and Phase One-s published figure was also around 12 EV (12 bits), the IQ-250 and IQ-350 supposedly deliver 14 EV corresponding to 14 bits of data and according to Phase One the camera has 14 bit wide data path.

Best regards
Erik

This is a great post. It helps me understand what issues really contribute to and control noise in sensor design. It's really very similar to the relationship of analog s/n to bit depth in audio recording.

 

[jerome_m]

There are a couple of points missing.

My post was indeed vastly simplified. I was just answering a specific question.

One is that modern CMOS sensors seem to have increased in full well capacity. I don't have actual numbers but I would guess that say Nikon D810 and say Phase One IQ have similar FWC (number of electron charges stored).

There is no indication that FWC has increased in a meaningful manner, especially when the pixels have be shrunk. Besides, increased FWC would only be useful if we were able to get more photons, which means either lower ISO or higher quantum efficiency or both. We are already relatively close to the maximum of quantum efficiency on modern cameras.

The other point is that the maximum meaningful amount of information from a sensor is FWC / readout noise. Readout noise for a Phase One IQ 180 may be around 12 electron charges. So the amount of information is really around 1:3800, say 1:4000. To represent that range you need 12 bits. So the output from the ADC is 16 bit but 4 bits of that is noise.

Yes and no. Having 4 bits of info on the noise will allow noise reduction to work better.

With recent CMOS sensors Sony and other vendors moved the ADCs on the chip. Normally there is an ADC for each column, so a normal 24 MP sensor has 6000 ADCs. Those ADCs are probably simple ramp type devices and can have long conversion times as each ADC only handles 4000 pixels.

That is true and is a Sony Patent.

Noise levels of on sensor ADCs may go down to 2 electron charges.

Yes and no. Part of that extremely low figure is due to correlated dual sampling, which adds some problem of its own. But discussing this would lead us too far from the title of the thread, which concerns the 645Z and the A7RII. The two cameras use a Sony chip.

 

[ErikKaffehr]

Comments, see below:

Best regards
Erik

My post was indeed vastly simplified. I was just answering a specific question.



There is no indication that FWC has increased in a meaningful manner, especially when the pixels have be shrunk. Besides, increased FWC would only be useful if we were able to get more photons, which means either lower ISO or higher quantum efficiency or both. We are already relatively close to the maximum of quantum efficiency on modern cameras.

Yes, but there are indications that minimum ISO is going down. Also, SNR seems to go up a bit. What I see is that the pixels are shrunk but FWC seems to stay around 60000.

For instance: Base ISO on the Sony Alpha is 119 ISO according to DxO and SNR at base ISO is 41.2 dB while the Sony A7rhas a base ISO 73 and SNR 45.2 dB. So SNR (18%), which is only function of full well capacity, has increased 4db. SNR is proportional to SQRT(FWC) so I would suggest that FWC/sensor area has been dubbled between the A900 and the A7r although the A7r has smaller pixels.

Sony Alpha 900 versus Sony A7R - Side by side camera comparison - DxOMark

Yes and no. Having 4 bits of info on the noise will allow noise reduction to work better.

I have not seen article confirming this, you perhaps have a reference?



That is true and is a Sony Patent.
Hi, I don't know if it is a Sony patent, I much doubt it because I think there are older implementations. Leica CMOSIS sensor also has column converters, but with higher base ISO and lower SNR. Toshiba has also column ADCs, so they may be more of a norm than an exception.



Yes and no. Part of that extremely low figure is due to correlated dual sampling, which adds some problem of its own.

Hi, correlated double sampling is simply measuring cell voltage before and after exposure, AFAIK, correct me if I am wrong.

Check this article: https://en.wikipedia.org/wiki/Correlated_double_sampling


But discussing this would lead us too far from the title of the thread, which concerns the 645Z and the A7RII. The two cameras use a Sony chip.

Quite right, both are 14 bit devices.