Erwin Puts: The Leica M9: part 5: M8/9 noise and dynamic range

[futureancient]

Advertisement (gone after registration)

All this is, is a sensor analysis on a pixel by pixel basis... it's hardly depressing news if he finds it's the same as the M8, it's near as damn it the same sensor, just bigger. Surely this is not what readers are that interested in though...is it ?

 

Shouldn't he be comparing images sized the same? I'd like to see what the results would be then... personally, I'm finding a big improvement over like-for-like pictures, and that, for me personally is all that counts.

 

Do you want to get hung up over the hardware minutiae, or do you want improvements in images / prints?

[Advertisement]

The Leica Q3 is available for pre-order at B&H Photo Video and Adorama

[t024484]

Hans,

 

What are your conclusions about the camera? How much differ from those of Puts?

 

First see this posting.

 

http://www.l-camera-forum.com/leica-forum/leica-m9-forum/101279-compressed-versus-uncompressed.html#post1069298

 

I have combined Erwins M9 figures with my results in one table.

My figures are for RAW, compressed aswell as uncompressed, and taken from green pixels.

At the point where S/N is a factor 10, a bit to my surprise we do not differ that much, although his graphs are weird.

But at S/N of 1, we have completely different figures.

 

Hans

 

P.S the figures 1.0, 0.5, 0.25 and 0.1 below mean a range from the highest S/N up to a S/N of resp 1.0, 2.0, 4.0 and 10.0 or 0, 1, 2 and 3.3 EV, .

 

 

 

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

[Guest EarlBurrellPhoto]

 

Do you want to get hung up over the hardware minutiae, or do you want improvements in images / prints?

 

Guaging by the number of threads doing the former I think the question answers itself, at least where Leica-related internet fora are concerned

[rosuna]

First see this posting.

 

http://www.l-camera-forum.com/leica-forum/leica-m9-forum/101279-compressed-versus-uncompressed.html#post1069298

 

I have combined Erwins M9 figures with my results in one table.

My figures are for RAW, compressed aswell as uncompressed, and taken from green pixels.

At the point where S/N is a factor 10, a bit to my surprise we do not differ that much, although his graphs are weird.

But at S/N of 1, we have completely different figures.

 

Hans

 

P.S the figures 1.0, 0.5, 0.25 and 0.1 below mean a range from the highest S/N up to a S/N of resp 1.0, 2.0, 4.0 and 10.0 or 0, 1, 2 and 3.3 EV, .

 

 

 

[ATTACH]167424[/ATTACH]

 

 

Thank you very much...

[sandymc]

I have combined Erwins M9 figures with my results in one table.

My figures are for RAW, compressed aswell as uncompressed, and taken from green pixels.

At the point where S/N is a factor 10, a bit to my surprise we do not differ that much, although his graphs are weird.

But at S/N of 1, we have completely different figures.

 

Hans

 

P.S the figures 1.0, 0.5, 0.25 and 0.1 below mean a range from the highest S/N up to a S/N of resp 1.0, 2.0, 4.0 and 10.0 or 0, 1, 2 and 3.3 EV, ]

 

For what its worth, the results I've had access to, and trust, are also different to Erwin's. They show the M9 marginally better than the M8 - maybe 0.1 to 0.25 stops better in terms of noise performance on a per-pixel level. Which translates to maybe 1 stop on an "equivalent sized print" basis.

 

But no question, the D3x significantly outperforms the M9 at high ISO.

 

Sandy

[futureancient]

Guaging by the number of threads doing the former I think the question answers itself, at least where Leica-related internet fora are concerned

 

 

You might well be right, frankly, it's boring as hell ! get on with taking superb pictures I even had someone come up to me on the street last week noticing I was using leica, he then saw it was an M9, - "oh but I heard they had software problems".. honestly, I couldn't quite believe what I was hearing.

[mjh]

Advertisement (gone after registration)

Do more channels at lower frequencies for extracting the signal have a substantial impact in noise and DR? I am referring to CCD sensors. This is the most important "improvement" of the S2 sensor compared to that of the DMR/M8/M9.

Lower read-out speeds and a correspondingly increased integration time does reduce read-out noise and the increase in sample size (14 rather than just 12 bits) does smooth out quantization errors; these are certainly important but gradual improvements.

 

The first CMOS sensor suitable for use in a DSLR was a major breakthrough – 9 years ago (Canon EOS D30). Since then, Canon has never stopped tweaking the sensor, improving the fill factor and the effectiveness of the microlenses, integrating more of the noise-suppression circuitry onto the chip, and so on, but again, these were evolutionary developments.

 

Anyway, even assuming that Puts' methodology is a bit strange, the broad picture from the comparison seems to be correct: CMOSes, whatever the techniques applied, have much better performance at high ISOs, even with smaller pixels...

I see no reason in principle why that should be the case, and I don’t see such a difference between CCD and CMOS in practice. For example, the Pentax K-7 and Sony Alpha380 have rather similar sensors (5.0 µm vs. 5.1 µm, 14.6 MP vs. 14.2 MP) despite the Samsung sensor in the K-7 being CMOS and the Alpha380 using a CCD; still the raw data of the Sony has lower noise at higher ISO levels (> 800).

[Angelos Viskadourakis]

well it seems that the majority is quite happy to pay 4500 euros and 5500 euros -total 10000 euros- for the amount and level and depth of changes between M8 and M9 in less than 2.5 years,after all is only 4000 euros per year.If we , LEICA M users are mostly happy with that ,thats fine with me,and i will try to forget that we all bought by thousands the M8s as a system that will follow the evolution of the M system as it was promised That is anyway the only reason for that kind of not so artistic tests and analysis.As we all know we are happy anyway with LEICA products and the principles that the company is following but some criticism when comes to who is paying what it helps to keep this highly and blindly world of profit only in a kind of respectfulll attitude.

Is the photographer not the camera that creates the value of the image but is not bad to discuss where the money of the photographer that supports a company goes

In my opinion ERWIN PUT is doing an excellent effort ,he offers alot in the community and as i said earlier when i tested my M8 two years ago against the CANON 5D the IMATEST gave exactly the same results as ERWINS PUT,i didnt like them then but i fully confirm them in photo action,of course all tests are nt dead accurate and as i said the best use of IMATEST is to compare between many cameras under the same conditions and evaluate the results,is not for a single camera evaluation but for comparison of various cameras the same time.Not like to be deatailed in tech talk but if a photographer doesnt know the dynamic range of his camera precisely how the hec is taking photos ,unless we are talking about point and shoot photography and we dont care about the tonality of the image.TIP:TRY THE M8 IN CONTRAST:MEDIUM LOW or LOW AND WILL GET SURPRISES IN THE POST PROCESS. .

[rosuna]

I see no reason in principle why that should be the case, and I don’t see such a difference between CCD and CMOS in practice. For example, the Pentax K-7 and Sony Alpha380 have rather similar sensors (5.0 µm vs. 5.1 µm, 14.6 MP vs. 14.2 MP) despite the Samsung sensor in the K-7 being CMOS and the Alpha380 using a CCD; still the raw data of the Sony has lower noise at higher ISO levels (> 800).

 

Well, CCDs are passive devices and CMOSes are more active (with tons of circuitry integrated into the chip, and handling signal processing at the pixel level). CMOSes seem to be more complex and the variance among different designs much larger. I was thinking on top class units. Moreover, CMOS evolve, but at sustained pace. Every generation of state of the art designs are better than the previous one, in one or more parameters (image size, DR, noise, tonal detail, live view support, now video...). The accumulated changes in a shot period of time (a few years) are impressive.

[devils-advocate]

Since when do we have to wade through waist-deep wallows of mathematical bullshit to find out if our camera takes good pictures? We didn't with film and we don't have to now. The M9's IQ is precisely equal to what comes out the other end after you take a picture, process it and print it. If it's good enough for you then it's good enough. If it isn't, it isn't.

 

Besides, none of us shoot with Leicas because of the world-beating IQ. Equal or better IQ is available for most applications through less expensive or more convenient equipment. We shoot Leica because we love the phyical experience of doing so. The fact that this beloved process can now yield really nice pictures is a nice bonus.

 

I love Erwin, but I kind of wish he'd take up golf.

[rosuna]

Besides, none of us shoot with Leicas because of the world-beating IQ.

 

In the film days that was the case...

[devils-advocate]

In the film days that was the case...

 

Agreed.

[sbelyaev]

Who cares whether M8/M9 have 6 or 7, or even 11, stops of DR and somewhat lower DR when ISO hits 1250? With introduction of Nikon D3s, Leica's sensor looks inadequate.

 

Is far as I concern 6 stops is good enough (paper can not handle higher contrast range) , as long as these 6 stops are of high quality. I don't like the fact that minimal overexposure (0.5-1 stop) washes out colors to the point that one can not properly recover them in a RAW converter (I don't even mention JPEGs). Nikon's ADR can handle this problem (to some extent).

 

In my opinion, software/firmware is the weakest part of Leica M system.

[tgray]

Is far as I concern 6 stops is good enough (paper can not handle higher contrast range)

 

I'm not attacking you, but I never understood this logic. You capture higher dynamic range, and compress it before putting it to paper. That's how film worked and that's how digital works. Heck, B&W negative film usually only has about 6 stops of range between film+base fog and the highlights, but that's just the contrast range of the actual negative. You can certainly capture more scene dynamic range, it just gets compressed on the film. That range is then well matched to the paper you print on, and we are all happy.

 

One of the icky things I find in digital is how highlights just go white abruptly. Higher dynamic range would help that, even if I then chose to roll them off gradually into white in post processing. Things would look more natural then to me.

 

Of course if we get too extreme with the dynamic range we can capture, we'll have to select a limited range for what goes down on paper, other wise everything will look like HDR. And we don't want that.

[t024484]

The discussion here is wether future improvements have to be expected from hardware or software.

For that reason I have update my M9 animation model with the read noise figure of the D3X, which is 1/3 of the M9 read noise. Everything else is kept unchanged.

The outcome of this is that if the M9 would have the same read noise as the D3X, the ISO 2500 performance would be similar.

 

The figures that Erwin Put gave us for the D3X at ISO 2500 were resp 5.63, 6.36, 7.56 and 9.88 for the ratios 1, 0.5, 0.25 and 0.1.

The step sizes though are a bit strange.and 5.63 and 9.88 are out of line.

The M9 with improved sensor noise would have 4.3, 6.3, 7.7 and 8.9, which would be a huge improvement and almost comparable to the D3X.

 

So I think that future software updates cannot solve the real problem, which is the read noise.

I have no idea if this has to do with the quality of the analog amplifier, with digital noise injection in the analog circuit, or with layout problems. But it is obvious that a sensor that produces less read noise would make high ISO values within reach.

 

Hans

 

 

M9 versus M9 with 1/3 read noise

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

[eronald]

Overall aren't the quality of the images on both cameras extemely high. Doesn't one need to look at other criteria for the final judgment as to purchase. Nikon v. Leica is like choosing a sports car when it comes to speed. How often do you go above a certain speed (ISO 400).

 

I shoot the D3x at 1600 ISO systematically, all the time, and usually print to A2 from a crop of the frame. There is some noise, but it sort of disappears when you look at the prints.

 

My Leica M8 has very good texture, but it is harder to use partly because light for fast shutter speeds runs out faster. 1600 ISO is not going to make me very happy here, 640 seems to be the real-life limit. I really think the M9 deserves a better sensor - the same problem will hit the S2.

 

Edmund

[delander †]

Hans is read noise the same as shot noise? I'm confused, how many types of noise are there?

 

Jeff

[markgay]

Quoting Michael J. Hußmann: Contrary to popular belief there weren’t any real breakthroughs in sensor technology these past years. There was some gradual improvement to be sure, but most advances have come from software

Re A.N.Other who called for CMOS sensors.

 

As Michael writes, there is a fond belief that sensor technology must have advanced, without any evidence. By far the greater advance, as reading DPReview will show, has come from in-camera image processing by software.

 

Secondly, why do people go on and on about high ISO performance? Can you imagine, in the days of film, people judging cameras exclusively by how they performed with ASA’s above 1600? With respect, the trend betrays ignorance that performance at low ISO ALSO varies between cameras. And given that the vast majority of photographs are taken in normal light, low ISO performance is more important in most cases.

 

The high ISO fetish most likely exists for several reasons. One of them – just look at the pictures you took as a teenager or a struggling young parent: Lots of shots around the table, in the living room, in bars, in clubs. It sure is nice to get clear pictures for once. In film days, it was impossible unless you carried a lighting kit and then someone else got the girls while you were packing away the strobes.

 

Another, its easier on the Internet to see the globules produced by noise at high ISO than it is to measure the truly beautiful, three dimensional clarity of Leica lenses in prints at moderate ISO. The high ISO fetish would not exist without jpegs on the Internet, so there's no point in frying Puts for not showing you Raw.

 

Thirdly, Leica excels in glass. Not just to let you shoot in low light but allowing you to get superior clarity and rendition in ANY light. At the DINs most of us shoot most of the time, the CCD sensor rends better images. CMOS has huge issues. If you doubt, read about moving picture cameras which put the sensor under much greater stress than still images.

 

Regards,

Mark

[Advertisement]

Hallo,

Look here: Leica M

[delander †]

I shoot the D3x at 1600 ISO systematically, all the time, and usually print to A2 from a crop of the frame. There is some noise, but it sort of disappears when you look at the prints.

 

My Leica M8 has very good texture, but it is harder to use partly because light for fast shutter speeds runs out faster. 1600 ISO is not going to make me very happy here, 640 seems to be the real-life limit. I really think the M9 deserves a better sensor - the same problem will hit the S2.

 

Edmund

 

But as all digital MF cameras have CCD sensors dont they all 'suffer' lack of high ISO perforamce compared to the latest CMOS cameras, and why do manufacturers and photographers keep using CCD?

 

Jeff

[markgay]

Why do people keep using CCD? Because it's not black and white.

 

The views of a manufacturer: CCD vs. CMOS

 

CCD: lower system noise and higher dynamic range.

CMOS: lower power consumption, better integration with in camera image processing.

 

Regards,

Mark