[biz-engineer]

When comparing Pentax K1 and 645z sensor image quality based on DXO and photonstophotos.net measurements, it looks like Pentax K1 is almost as good as the 645z in terms of S/N, DR etc.. for shadow recovery and tones.
However, I have raw files from both the K1 (my main camera) and 645z DNG downloads, and they behave rather differently in the raw converter.

The DNG from the 645z have a lot of room before turning bad when moving the sliders, feels like big cylinder "diesel" engine.
The DNG from the K1 turn bad much more quickly when pushing sliders. From the DXO and photonstophoto curves, there should be only about 0.3ev difference of DR between those sensors.
In the RAW processor, it feel like more than a stop of headroom, and smooth tonalities from the 645z files.
K1 files deteriorate quickly with higher ISO, 645z not so much. Where's the magic?

Where does that difference come from?

[RICHARD L.]

Indeed, the K1, coupled with a very good lens (here the DFA* 70-200 mm f/2.8), produces images reaching the digital medium-format field.

[btnapa]

K1 files deteriorate quickly with higher ISO, 645z not so much. Where's the magic?

Interesting question. I would like to know the answer too. Perhaps someone who owns both and has shot the same scene with both cameras can shed some light.

[zjacreman]

Where does that difference come from?

Bigger photosites tend to make for better results. The 645z has more pixels, but a lot more area, so it has a larger pixel pitch and 20% bigger photo sites.

Pentax K-1 vs. Pentax 645Z - Sensor Comparison

[BarryE]

Maybe DXO don't know what they are doing ;-) ? However, your questions will create comment. I've attempted what you've tried and the results are not quite the same. I concluded that, sometimes, 645Z have more adjustment headroom, but often, like DXO show, the difference is not always that marked. It probably depends on usage and style...

There will be owners of 645Zs who will say that the 'sometimes' should be 'very nearly always' or 'always'. As they've had more time to use their 645Zs their views are likely to be accurate, but I guess you have to take into account a built in bias that can come from making that medium format investment;-)

I'll be interested to see what's said ...

[biz-engineer]

As they've had more time to use their 645Zs their views are likely to be accurate, but I guess you have to take into account a built in bias that can come from making that medium format investment;-)

I haven't made that 645z investment ;-) I have a few 645z raw files underexposed, if I correct exposure there is not visible degradation of tonality and I feel like there is a log curve behind my sliders, meaning I have to push real hard to cripple the image. With my K1 files, there is no need to push sliders to the max to see colors crippled. Is there something in DNG files that tell the raw converter how sensitive sliders are to the raw data?.

---------- Post added 16-02-21 at 18:18 ----------

The 645z has more pixels, but a lot more area, so it has a larger pixel pitch and 20% bigger photo sites.

According to DXO , 1 pixel of a K1 performs about the same as 1 pixel of a 645z. That would mean, 645z noise equivalent sensor size could be about 80% of 1452m2 (1161m2 equ.) since 645z pixels have an area 20% larger with same S/N. 1161m2 (Z) vs 861 (K1) corresponds to about 0.3ev better for the 645z, not a big difference. From theory, I could dial +0.3ev of exposure compensation on the K1 and not see any difference with 645z, but I don't think that would ever close the gap I can see in raw developer.

[JensE]

If you compare the DPreview test scene for both, the K-1 is close only when using PixelShift, otherwise there is clearly less than a stop difference at higher ISO, but maybe closer to 1 than 0 stops. Comparing 'normal' mode at base ISO, the 645Z looks a lot cleaner and it takes almost 2 stops ...
I'd recommend to explore different parts of the scene and play with light/dark and full/comp display, maybe at different ISO.

---

Last edited by JensE; 02-16-2021 at 10:43 AM.

[UncleVanya]

You might want to look at photons to photos site.

Photographic Dynamic Range versus ISO Setting

They also show only about 0.35 stops of dynamic range difference, but the low light EV shows an advantage for the 645z of about 0.54 EV in low light.

If you are interested in the testing you can talk to the site owner and run your own tests on your own equipment.

[BarryE]

If you compare the DPreview test scene for both, the K-1 is close only when using PixelShift, otherwise there is clearly less than a stop difference, but maybe closer to 1 than 0 stops. I'd recommend to explore different parts of the scene and play with light/dark and full/comp display, maybe at different ISO.

I think this is the 'sometimes' I referred to. Sometimes being in different areas of a photograph as well as 'sometimes' being from a different photo. If I'm in a situation where I know the DR of a K-1 is at its limit then I bracket and a simple blend in PS and I have all I need. A 645Z would be able to capture the scene, sometimes, in one go, but bracketing is hardly tricky - and the kit weighs a little less too.

I'm not trying to say there's little difference between the two cameras, just that the differences may not always be as great as is assumed when mitigating procedures are considered - at low ISO, say below 800. My experience of the RAW files from a 645Z is limited, but I've done several test shot at shows and download comparisons.

[biz-engineer]

If you compare the DPreview test scene for both, the K-1 is close only when using PixelShift

Yep, pixel shift shift easily closes the gap for the noise, although PS isn't always usable.

[SFTphotography]

I run a D810 and 645z. For all intensive purposes the D810 and K1 are identical sensors and perform very similarly.


What I find is that if you pull back the shadows at base ISO the 645z and D810 perform pretty similarly. Where the 645z is cleaner is in the highlights - they just recover more naturally. Still £ for £ there is no 50mp DSLR that touches the 645z for resolution and DR. That said I tend to ETTR where possible and never have to pull too much shadow back. I shoot exclusively landscape and use grad filters to ensure I can comfortably expose the scene. Anyway - a few blacks ain't bad in the frame. You don't have to see everything.


The 850 gets close but is still short 5mp.That 36mp sensor in the K1 and D810 is glorious though - and you can get used D810's for £800 or so now. An incredible camera for the money. K1's are still a bit rare on the ground and 36mp seems a little low - sooner they fit that with the D850 sensor or even the 61mp sensor off the A7R4 the better.

[photoptimist]

As others have said, the larger pixels of the 645Z reduce the pixel-peeper noise level and, especially, reduce the picture-level noise level.

I'd also wonder if the 645Z has better noise structure. One of the huge technological improvements in digital cameras was a trend toward more film-like random noise. Early digital cameras tended to have very structured noise such as vertical or horizontal bands of noise or color artifacts. Using high ISO or pushing the shadows would make very ugly banding appear.

Improvements in semiconductor manufacturing, EMI/RFI control, and corrections for row and column black-levels seem to have helped a lot.

Two cameras could have nearly identical RMS noise levels and DR estimates but if one camera has truly random grain and the other has banded grain, the random-grain camera would have much nicer high-ISO and pushed images.

[AstroDave]

You don't have to see everything.

Love it! A quote for the ages!

[biz-engineer]

Two cameras could have nearly identical RMS noise levels and DR estimates but if one camera has truly random grain and the other has banded grain, the random-grain camera would have much nicer high-ISO and pushed images.

I'm question the quality of LSBs of the 14bits A/D in small sensors vs larger sensors. In smaller sensors the A/D can be 14bits but more like 12bits effective and the 2 last bits mostly random values. I was wondering about linearity, how linear is the A/D in the 645z sensor vs K1 sensor? Could be that the K1 sensor has 14bit DR on paper, but not able to maintain the color mix at low light levels due to gain mismatches between primary channels varying from sensor to sensor?

[JensE]

Could be that the K1 sensor has 14bit DR on paper, but not able to maintain the color mix at low light levels due to gain mismatches between primary channels varying from sensor to sensor?

I don't think so. That would cause a systematic color cast in dark areas, which at least I haven't observed.