5DMK2 vs K20D high ISO comparison

Note that I also did a fair amount of testing (mostly comparing lenses or cameras I have in the focal lenghts which overlap) , it really does take time, and conclusions are always a bit subjective ... thats where a good theory is comforting : at least the laws of physics don't change ;-)

Kind regards, Pieter

pcarfan said:

Gordon,

Given the same dynamic range for a given iso, can two different
models choose to have better shadow details than the other ? if yes,
is it by sacrificing highlight detail ?

Click to expand...

I think you mean two different models for two different manufacturers. No, if the DR is the same as per DxO's raw tests, then the deep shadow noise will be the same from each; the only potential difference is the DxO's definition of "real" ISO defined by the exposure range of the raw data could be different. In that case, the camera with higher "real" ISO would have more latitude to Expose To The Right (ETTR) and thus minimize noise in the shadows that way.

For example, suppose we have two cameras with the same measured DR as per their manufacturer stated ISO but camera A is really at ISO 80 where camera B is at ISO 100. Now in order to get the same histogram for camera A we actually have to expose a little more, which will shift the same dim shadow detail brightness away from the noise floor which is really the measured DR level and this may make the push processed images from camera A look cleaner. This isn't that surprising in that generally lower ISO's have less noise.

The newer Nikon models like the D90 and D3X seem to be using some sensor trickery in order to push the electron capacity up to unprecedented levels per unit area, also have tuned their active reset circuitry for low read noise, and obviously have very good digital converters, as compared to say Canon even with the 5DM2.

Regards, GordonBGood

tigrebleu said:

GordonBGood said:

tigrebleu said:

Also, Pentax has an off-hands approach to noise, unlike Canon, which
uses some NR on its RAW files (before in-camera processing) from ISO
800, if I remember correctly.

Click to expand...

I'm not so sure that Pentax is the one that still has the hands off
approach to noise, at least any more. Starting with the K20D there
were some pretty definite signs of a little bit of noise reduction to
the raw data files for ISO 1600 and above

Click to expand...

Yup, even DPR said so in their review of the K20D. I forgot to
mention that I was talking about the entry-level through
advanced-amateur-level DSLRs. I suspect the Canon 5D and Nikon D700
FF DSLRs are closer to pro DSLRs than they are to amateur APS-C
DSLRs, at least in terms of how noise is managed.

Click to expand...

Is the Canon 1000D entry enough for you. Even this camera shows no signs of noise reduction in raw, although it does show the signs of a poor digital converter limited its DR at low ISO's.

tigrebleu said:

tigrebleu said:

I've never seen any signs of noise suppression for raw data in Canon
raw files, to the extent that they don't even offset the data so
black is zero.

Click to expand...

A friend of mine who owns a 5DMkII suspects his camera applies NR to
the RAW files from ISO 800. But he admits he's no expert, and also
says it could be just a matter of the circumstances in which the
picture was taken (darker or brighter subject, background, can make
noise more or less obvious to the naked eye).

Click to expand...

The only real way to test this is the DxO way from raw data and a step chart image. There are too many variable in other types of "gathering impressions".

tigrebleu said:

tigrebleu said:

I think we have to be men enough to admit that Canon truly has
excellent high ISO results with their multi generation CMOS sensor
designs.

Click to expand...

Canon used to be the leader in NR technology. Now Nikon took over
that role IMHO. The D3 and D700 just have DR and noise levels out of
this world. Clean ISO 6400 is just something an APS-C sensor isn't
capable of now (it might be possible in a few months though:
technology evolves so rapidly).

Click to expand...

Canon still has the best technology to produce the maximum DR from their sensors ; however they are let down by their digital converter that limits the DR at low ISO's. For the 5DM2, if it had a good enough digital converter, it would have a DR at low ISO of almost a stop better than the Nikon D3X, the current king. This has nothing to do with NR but with designing the sensor and digital converter to be very low noise in the first place, which Canon have done for their sensor but haven't been so successful for the converter. The Nikon D3 and D700 have an excellent sensor but no where near the quality of converter as the D3X or even the D90 for that matter.

Regards, GordonBGood

samusan said:

samusan said:

It's just that most FF models have larger pixels than aps-c sensors
which provides for less noise at higher iso.

Click to expand...

The sensor is not what collects light. The lens does the gathering,
and projects it onto the sensor. Projecting onto a smaller or bigger
surface is no big deal, FF sensors are not what makes FF systems good
with low light.

Click to expand...

I think your missing his point. This has been an issue in astro imaging ever since CCD imagers have been used. The area of the pixel determines how many actual photons it can capture before blooming or bleedover occurs. When you go to a higher ISO you are just turning the pixel's amplifier up, i.e. multiplying the number of photons the pixel is registering as hits. It works till you start to saturate the maximum number it can handle then noise starts interfering with the result. this is the effect of photo saturation or blooming. A pixel with a larger area can capture more photons or amplified photons before it has a problem. Also the bane of CCD/CMOS is heat. The longer it is on the hotter it gets and you start getting stuck pixels or hot pixels as they call them in DSLRs.

If a FF camera had the same pixel density as the K20D, about 34 MP, it would have exactly the issues as the K20D. The lenses have no part in this at all. At the same aperture (real aperture) the same number of photons will fall on each square unit of the sensor regardless of the type of lens, vignetting aside, or the type of camera. At F1.4 or any other aperture a lens deposits the same number of photons per square unit regardless of the overall physical size of that sensor.

The reason they work better is that they have pixels about the size of an 8 MP APS-C camera sensor and the technology has improved since that size has come out. All this is independent of the rest of the camera. The reason why denser and denser sensor arrays are actually producing better and better results most of the time instead of worse is the fact that the technology of controlling the output, Bayer interpolation, amplifier gain, and noise reduction is improving as well.
Kent Gittings

Some lenses are darker than others due to groups, elements, coating, etc.

Same Sensitivity, same WB, same Av, same Tv, does not guarantee the same histogram at all, unless the same lens is used.

Since it's an ISO test, not a resolution test - it may make more sense to use the same 3rd-party lens model with different mounts, instead of the sharpest lens you can find. Has that been tried?

leeak said:

According to Dxomark.com, 5DMkII has a score or 79 and the GX20 and
K20D have scores of 68.7 and 66 respectively.

They stated that "DxOMark Sensor scale is logarithmic and a 5-point
difference on the scale corresponds to a gain or loss of sensitivity
of 1/3 of a stop. We nevertheless report DxOMark Sensor Scale test
results to one decimal place to avoid the clustering of data that
would occur with any arbitrary algebraic rounding."

As a result, the 5DMkII scores 10-13 points more than the GX20 and
K20D which suggest an improvement of less than one stop. So my
question is, how do we explain the difference in Brad's observation
and DxOMark results?

Click to expand...

DxO's scores are overall scores including low ISO DR, high ISO usability, and something called "Color Depth". Brad is only comparing high ISO performance.

Regards, GordonBGood

KentG said:

KentG said:

KentG said:

It's just that most FF models have larger pixels than aps-c sensors
which provides for less noise at higher iso.

Click to expand...

The sensor is not what collects light. The lens does the gathering,
and projects it onto the sensor. Projecting onto a smaller or bigger
surface is no big deal, FF sensors are not what makes FF systems good
with low light.

Click to expand...

I think your missing his point. This has been an issue in astro
imaging ever since CCD imagers have been used. The area of the pixel
determines how many actual photons it can capture before blooming or
bleedover occurs. When you go to a higher ISO you are just turning

Click to expand...

Lots of details that I mostly agree with.

But have you thought about how using 4 smaller pixels instead of one on the same surface changes things wrt noise in the final image ? Basically, it doesn't.

So if you look at the final image (not a 1000x1000 crop or any other kind of pixel peeping), you don't care whether the photons collected by the lens are projected onto 10Mpix or 30...

That's why DxOMark shows 2 sets of measures :
-"print" stats (i.e. normalized final size, =quality of the image as a whole)
-"screen" stats (representative of the feeling you'd get when pixel-peeping)

what you say about bigger pixels only concern "screen stats", while I'm interested in "print" stats.

KentG said:

Also the bane of CCD/CMOS is heat. The
longer it is on the hotter it gets and you start getting stuck pixels
or hot pixels as they call them in DSLRs.

Click to expand...

But really, why complicate things with exposure length ? it's yet another parameter, that has nothing to do with this thread.

If you do consider heating, remember bigger chips heat more, the advantage when looking at long exposures is not where you think it is.

KentG said:

If a FF camera had the same pixel density as the K20D, about 34 MP,
it would have exactly the issues as the K20D. The lenses have no part
in this at all. At the same aperture (real aperture) the same number
of photons will fall on each square unit of the sensor regardless of

Click to expand...

No, the "real aperture" is a diameter length. At same real aperture, same FoV, distance to subject, and different crop-factors, you get the same number of photons in total , not per surface.
To get the same number of photons per surface you must use the same f-ratio .

KentG said:

the type of lens, vignetting aside, or the type of camera. At F1.4 or
any other aperture a lens deposits the same number of photons per
square unit regardless of the overall physical size of that sensor.

Click to expand...

for the one same lens at same focal length, true. But the point is, you use a different lens / different focal length to get the same image.

KentG said:

The reason they work better is that they have pixels about the size
of an 8 MP APS-C camera sensor and the technology has improved since
that size has come out. All this is independent of the rest of the
camera.

Click to expand...

It's exactly the other way round.

Compare (for the same scene, same FoV, same distance to subject, same exposure length) a FA50mm f/1.4 on FF

to a DA 33.3mm f/0.93 on a 1.5x crop, you get the same image, same shot-noise, same diffraction effects on final image, just as long as both sensors have adequately enough pixels, and somewhat similar read-noise / AD converters.

So, I insist, the biggest advantage of FF systems is their lenses. In similar situations, they offer much wider apertures (in mm), and that's the whole point, the image ends up with far less shot-noise (and possibly a bit less, or a bit more, read-noise. I think it is now a bit less, but a few years back it was a bit more)

Here the OP compares two systems :
50mm/1.4 => 35.7mm
31mm/1.8 => 17.2mm
That's 2.07x wider aperture if you go for the maximum of each system.

The FF here could cope with 2.1 stops darker scene than the APS-C can with the same shot noise (but at a shallower DoF, as everyone already agreed).

The difference in read-noise from both sensors is bound to be fairly negelectable (say, + - 0.5 stops ?) compared to the 2.1 stops more light advantage coming from the FF lens !

Sure the sensor does cause some difference via the read noise. It's there, but it should not be confused with the huge advantage in shot-noise that stems from the lens .
--
Samusan

samusan said:

You might not have noticed part of my post but your results don't
show a lesser sensitivity from the K20D.
31mm differs from 50/1.5. The difference results in 0.21 stops less
light per surface unit...

so you underexposed the K20D compared to the Canon.

Click to expand...

BTW, to correct that mismatch you can push all K20D raws by +0.21ev and adjust the ISO labels accordingly :
ISO 1600 should be labeled 1850
ISO 3200 should be labeled 3700
ISO 6400 should be labeled 7400

-- Sam

wy2lam said:

Some lenses are darker than others due to groups, elements, coating,
etc.

Same Sensitivity, same WB, same Av, same Tv, does not guarantee the
same histogram at all, unless the same lens is used.

Since it's an ISO test, not a resolution test - it may make more
sense to use the same 3rd-party lens model with different mounts,
instead of the sharpest lens you can find. Has that been tried?

Click to expand...

Could be the lens is producing the different sensitivities, yes. I don't have the equipment to test that idea though.

Which, maybe, doesn't help with the high ISO noise performance.

No surprise, though: Pentax metering has a tendency to expose for the highlights, which tends to prevent highlight clipping, but which in turn result in darker images in which the noise can be more visible to the eye.

Also, DxO measurements show that the mesured ISO values are higher on the Canon 5DMkII's than they are on the Pentax K20D (DxO measures for ISO 1600 are 1383 for the 5D and 1093 for the K20D). That's more than a slight difference and I guess it would truly help the K20D if it was used with a slight overexposure to compensate that ISO 290 difference in true ISO value with the 5DMkII.

I think the difference between ISO 1093 and ISO 1383 is about a third of a stop... Or am I wrong?

I'm not saying that changing the K20D's exposure will allow it to compete with the 5DMkII: the FF sensor is just much better than the APS-C in terms of noise performance.

But a slight +0.3 or +0.5 EV exposure compensation on the K20D might show the Pentax flagship isn't as bas as it looks like, even if such an exposure compensation won't suddenly make it a good high ISO performer.

--
Once you've mastered the technique and the equipment, you can concentrate on
the more important aspects of photography: originality, atmosphere, emotion
and — ultimately — soul.
— Jeff

GordonBGood said:

tigrebleu said:

Yup, even DPR said so in their review of the K20D. I forgot to
mention that I was talking about the entry-level through
advanced-amateur-level DSLRs. I suspect the Canon 5D and Nikon D700
FF DSLRs are closer to pro DSLRs than they are to amateur APS-C
DSLRs, at least in terms of how noise is managed.

Click to expand...

Is the Canon 1000D entry enough for you. Even this camera shows no
signs of noise reduction in raw, although it does show the signs of a
poor digital converter limited its DR at low ISO's.

Click to expand...

Yup, the 1000D is entry enough. No need to be sarcastic.


Well, I compared the DxOmarks readings for the EOS 1000D, the Nikon D90 and the Pentax K20D and I must agree.

Beyond the fact that the 1000D has better SNR 18% scores than the K20D, the graph is also much more linear with the Canon 1000D than it is with the K20D. The Pentax flagship has a slighlty curved graph (inwards, poorer SNR) when reaching higher ISO values, which would indicate some NR on the RAW files. Feel free to correct me if I'm wrong.

And although it doesn't look so bad on first glance, I remembered that these dB values were show on a logarithmic scale! So a slight curve is not so slight after all...

GordonBGood said:

GordonBGood said:

A friend of mine who owns a 5DMkII suspects his camera applies NR to
the RAW files from ISO 800. But he admits he's no expert, and also
says it could be just a matter of the circumstances in which the
picture was taken (darker or brighter subject, background, can make
noise more or less obvious to the naked eye).

Click to expand...

The only real way to test this is the DxO way from raw data and a
step chart image. There are too many variable in other types of
"gathering impressions".

Click to expand...

Agreed.

GordonBGood said:

GordonBGood said:

Canon used to be the leader in NR technology. Now Nikon took over
that role IMHO. The D3 and D700 just have DR and noise levels out of
this world. Clean ISO 6400 is just something an APS-C sensor isn't
capable of now (it might be possible in a few months though:
technology evolves so rapidly).

Click to expand...

Canon still has the best technology to produce the maximum DR from
their sensors ; however they are let down by their digital
converter that limits the DR at low ISO's. For the 5DM2, if it had a
good enough digital converter, it would have a DR at low ISO of
almost a stop better than the Nikon D3X, the current king. This has
nothing to do with NR but with designing the sensor and digital
converter to be very low noise in the first place, which Canon have
done for their sensor but haven't been so successful for the
converter. The Nikon D3 and D700 have an excellent sensor but no
where near the quality of converter as the D3X or even the D90 for
that matter.

Click to expand...

Well, that explains a lot of things. I knew the ADC was very important in the imaging pipeline when it comes to dynamic range, but I didn't know it had such an influence on noise level.

I mean, every piece of electronics that plays a role in the imaging pipeline has an effect on the overall noise levels, so I guess I should have thought about that: a poor quality ADC must have some effect on the noise levels. If they're very low at the sensor level, a poor quality ADC can still be quite alright in the end. But if the noise levels on the sensor are on the high side, a poor ADC will just make it worst.

BTW, what do you think Pentax needs to improve on its future K30D to have better noise performance? A higher quality ADC or a better sensor? Or both, maybe?

Thanks for posting. Your comments are always appreciated, Gordon.

--
Once you've mastered the technique and the equipment, you can concentrate on
the more important aspects of photography: originality, atmosphere, emotion
and — ultimately — soul.
— Jeff

tigrebleu said:

Which, maybe, doesn't help with the high ISO noise performance.

No surprise, though: Pentax metering has a tendency to expose for the
highlights, which tends to prevent highlight clipping, but which in
turn result in darker images in which the noise can be more visible
to the eye.

Also, DxO measurements show that the mesured ISO values are higher on
the Canon 5DMkII's than they are on the Pentax K20D (DxO measures for
ISO 1600 are 1383 for the 5D and 1093 for the K20D). That's more than
a slight difference and I guess it would truly help the K20D if it
was used with a slight overexposure to compensate that ISO 290
difference in true ISO value with the 5DMkII.

I think the difference between ISO 1093 and ISO 1383 is about a third
of a stop... Or am I wrong?

I'm not saying that changing the K20D's exposure will allow it to
compete with the 5DMkII: the FF sensor is just much better than the
APS-C in terms of noise performance.

But a slight +0.3 or +0.5 EV exposure compensation on the K20D might
show the Pentax flagship isn't as bas as it looks like, even if such
an exposure compensation won't suddenly make it a good high ISO
performer.

--
Once you've mastered the technique and the equipment, you can
concentrate on
the more important aspects of photography: originality, atmosphere,
emotion
and — ultimately — soul.
— Jeff

Click to expand...

I agree the K20D images are underexposed. All images were shot manual. Although your statement about Pentax metering philosophy is true, metering tendencies did not come into play here. Both cameras had the exact same exposure settings for each comparison shot, my original post in this thread described the method I used in more detail. I shot the 5D2 images first, and then used the same settings on the K20D without even looking at the LCD, didn't even realize the underexposure at the time. The sensitivity difference in the two cameras at the same stated ISO was an unexpected finding in the comparison.

To correctly expose the K20 in the ISO 1600 shot I would have had to go to 1/30 and F1.8. This isn't practical for what I usually photograph in the low light setting, which is candids, mostly of kids. I appreciate what you are saying, but I don't think the K20D did all that poorly. I've always considered ISO 1600 usable on the K20, which is pretty good for today's APS-C cameras.

I wouldn't normally post images like this because I know it stirs up bad sentiments, but I did it as a followup to someone else's claims in a different thread that they observed very little difference between these 2 cameras in low light/high ISO situations. My observations were different so I took a few shots to demonstrate the performance differences I observed and thought the fairest way of doing so would be to set both cameras with same exposure and ISO settings.

Brad Hollister said:

To correctly expose the K20 in the ISO 1600 shot I would have had to
go to 1/30 and F1.8. This isn't practical for what I usually
photograph in the low light setting, which is candids, mostly of
kids. I appreciate what you are saying, but I don't think the K20D
did all that poorly. I've always considered ISO 1600 usable on the
K20, which is pretty good for today's APS-C cameras.

Click to expand...

Of course the K20D isn't bad at all. It even tops my K10D by more than a bit at ISO 1600 and it's still usable at ISO 3200.

Overexposing the K20D's shots by about 0.3 EV would show how both cameras compare at near-identical true ISO reading, instead of the manufacturers' claimed ISO setting.

But of course, that would end in a slightly slower shutter speed for the K20D. But a 0.3 to 0.5 EV difference isn't always a problem and even at 1/30 and F/1.8, it would be enough for many low light situations.

I shoot stage performance at speeds between 1/10 and 1/30 (F/2.8-F/4) very often and I most often get very good results. The trick is to wait for the right moment, when the subject is rather steady but is still in motion. And slow shutter speeds have their advantage when it's about showing motion rather than freezing it.

--
Once you've mastered the technique and the equipment, you can concentrate on
the more important aspects of photography: originality, atmosphere, emotion
and — ultimately — soul.
— Jeff

tigrebleu said:

GordonBGood said:

Is the Canon 1000D entry enough for you? Even this camera shows no
signs of noise reduction in raw, although it does show the signs of a
poor digital converter limited its DR at low ISO's.

Click to expand...

Yup, the 1000D is entry enough. No need to be sarcastic.

Click to expand...

Sorry, I just took exception to the commonly held belief that Canon does raw noise reduction based on the misunderstanding that their multi generation active reset circuitry is the same kind of detail smearing noise reduction as is applied off sensor by others such as Sony (and now Pentax).

tigrebleu said:

Well, I compared the DxOmarks readings for the EOS 1000D, the Nikon
D90 and the Pentax K20D and I must agree.

Beyond the fact that the 1000D has better SNR 18% scores than the
K20D, the graph is also much more linear with the Canon 1000D than it
is with the K20D. The Pentax flagship has a slighlty curved graph
(inwards, poorer SNR) when reaching higher ISO values, which would
indicate some NR on the RAW files. Feel free to correct me if I'm
wrong.

Click to expand...

Actually, the brighter signal SNR's as for 18% and above aren't all that bad for the K20D considering the photosite density. The brighter signal SNR's are mostly governed by the electron well capacity and Samsung designed the K20D sensor to maximize this. Once we start considering photosite density, we should likely be comparing the K20D with the Canon 50D and the Nikon D90, as these are as closest each manufacturer has as far as density is concerned. Notice how the K20D has a SNR of 44.1 db at 100% brightness, where the Canon 50D has only 42.1 and the Nikon D90 has 45.3 (lower photosite density) after pulling every trick in the book to maximize electron well capacity. The K20D sensor does't have a problem with electron well capacity.

tigrebleu said:

Well, that explains a lot of things. I knew the ADC was very
important in the imaging pipeline when it comes to dynamic range, but
I didn't know it had such an influence on noise level.

Click to expand...

Black noise level, for which the ADC is one of the contributors, governs the "noise floor" which is what limits Dynamic Range (DR).

tigrebleu said:

I mean, every piece of electronics that plays a role in the imaging
pipeline has an effect on the overall noise levels, so I guess I
should have thought about that: a poor quality ADC must have some
effect on the noise levels. If they're very low at the sensor level,
a poor quality ADC can still be quite alright in the end. But if the
noise levels on the sensor are on the high side, a poor ADC will just
make it worst.

Click to expand...

In Canon's case, their poor quality ADC's are their main limit to DR in that the noise from their ADC's is the main contributor to the "noise floor" in the black levels. So in spite of having the most noise free sensors in the industry, they can't offer the widest DR. To see this, look at the DR chart on DxO for the Canon 50D and notice how the curve flattens out as the line goes from high to low ISO's. This flattening is caused by the ADC noise. If this flattening were not there, one can extrapolate that there would be a DR according to DxO's definition of over 13 Ev at ISO 100.

Looking at the same chart for the D90, notice how the graph doesn't flatten out nearly as much and the main limit to DR is that the base ISO is 200 and not 100. Extrapolation of this line to ISO 100 shows that the DR should be well over 12 Ev and even approaching 13 Ev if there were an ISO 100.

For the K20D, the DR chart is terrible, which we know is not due to the ADC (because the chart for the K200D, which AFAIK uses the same ADC doesn't have the flattening), but due to the on-sensor variable gain amplifier that Samsung designed into the sensor thinking it would improve black SNR when it actually made it worse.

tigrebleu said:

BTW, what do you think Pentax needs to improve on its future K30D to
have better noise performance? A higher quality ADC or a better
sensor? Or both, maybe?

Click to expand...

As I said, the ADC is actually pretty good as seen on the K200D, and Pentax/Samsung mostly need to make improvements to the CMOS sensor, removing the variable gain components which make the noise worse or redesigning them so they don't add this noise, as well as working on their active reset technology so the black noise from the sensor is about half to a quarter of what it is now. It isn't so surprising that some sensor improvements need to be made considering that this is a first generation design.

That said, many users never push process and thus will never use the full DR of their cameras above about 9 Ev in that they always Expose To The Right (ETTR). However, low sensor black noise (not the ADC noise) does have an effect on the amount of noise at high ISO's as these higher ISO gain factors mean that one is using a much lower range of the sensor, as in ISO 3200 only uses the range from 5 stops below full bright downward, thus the noise that would be in deep shadows at ISO 100 becomes the noise in the mid tones at ISO 3200.

Regards, GordonBGood

DxO Stated: Canon 5D II = 1093 (ISO 1600) and Pentax K20D: 1383 (ISO 1600).

They where switched in your post.

--
Bye4now



http://www.indots.nl

I have the deepest respect for all those people who like me.

Richard Smals said:

DxO Stated: Canon 5D II = 1093 (ISO 1600) and Pentax K20D: 1383 (ISO
1600).

They where switched in your post.

Click to expand...

So then the post above was worth nothing... Sorry.

--
Once you've mastered the technique and the equipment, you can concentrate on
the more important aspects of photography: originality, atmosphere, emotion
and — ultimately — soul.
— Jeff

Thanks a lot for taking the time. Always interesting to hear of the Pro's and Cons of different tools.

--
kind regards
Sune



Trusted Reviews : ”K20 most powerful semi-pro DSLR in the world'”