-
Welcome to the new forums! Please read this first. For known issues we are working to resolve, click here.
Is 16.7 MP a technical limitation or a marketing decision?
- Thread starter Bernard Languillier
- Start date
AlanG
Senior Member
I really am not following you at all. Of course if you crop in on a lens many designs are likely to have more even illumination regardless of whether film or digital is used. All wide angle designs have issues with even illumination. Some designs (mostly retrofocus) have solved this better than others. The very best and widest view camera lenses are not retrofocus and require center filters to get relatively even results. Some like the Biogons (for medium format and view cameras) have more even illumination but are not as wide angle as Super Angulons and Grandagons (105 to 120 degree angle of view.) And with some designs, the maximum angle of view is only achieved when stopped down pretty far.
Various designs require different distances from the film plane (or sensor) for a given focal length. Thus the angle of light they present to the sensor corners can vary widely. This has nothing to do with the size of the sensor.
The only way that the smaller sensor will have better results is if the wide angles designed for it have a better ability to evenly illuminate the sensor. If sensors require a design that somehow uses less of an angle to illuminate the edges, I don't see why this same design couldn't be used in a correspondingly longer focal length for a larger sensor.
Are you saying that there is something about the smaller sensor that makes it possible to use lens designs that won't work on a larger sensor? I don't see how that can be.
By the way. I use a 1Ds with the Canon 24 tilt shift and Nikkor 28 and 35 PC lenses. If anything the 24 TSE lens really can illuminate the sensor from extreme angles yet it works quite well. How do you explain this? It seems to me that this "edge issue" is a non issue with the Canon 1Ds. Have you tried one?
Alan Goldstein
http://www.goldsteinphoto.com
Bernard Languillier
Senior Member
I am not questioning your skills. I am just saying that the 30% you mention would mostly be influenced by:
- the Canon technology in terms of process,
- the experience of the people running it,...
This is assuming that it took Canon 2 years to design the new sensor of the 1Ds MKII, right? The fact is, even if it would probably have taken longer to design a 22 Mp sensor, we have zero information either way regarding whether it would have adveresely affected the release date of the camera, right?
Well, good for them, but as a customer, I don't care much about their business. All I feel is that I could maybe have been a potential buyer of a 22MPixel camera, but today it is only 16.7 MP...
I am aware that it is probably enough for many people, and that not only pixel matters etc...
Let's close this discussion anyway, I think that our respective opinions are clearly understood. Both are based on assumptions that we are not really in a position to proof.
Best regards,
Bernard
AlanG
Senior Member
--
Of course I'd like it. And I'd also like a much smaller charger and batteries with state of charge indicators on them. But I want them to work with my existing cameras. Of course I'll always carry extra batteries around. And extra cameras. It goes with being a pro. (Three cameras, four or five batteries, no worries.)
Alan Goldstein
http://www.goldsteinphoto.com
James L Wilson
Senior Member
Folks, this is likely the basis for why a manufacturer picks the imager dimensions for "what is the next step" in CMOS or CCD imager technology.
Truth is that CANON and SONY are capable of producing the 43MP imager shown below (today), however, it would only be affordable by NASA. The time and materials based progression of steps, is based on, lets say for example, Canon's ability to improve design, fabrication techniques, materials, and the spread of engineering costs over an increasing number of deliverable units.
Do the math, 11x17x300x300 = 16,800,000 approx. This is two 8.5x11" page centerfold, set by the USA print industry (medium format territory). The next pixel count steps are, from a "what are you going to do with the images" perspective:
1. 6"x8" print at 300 dpi yields 4,320,000 pixels
2. 8"x10" print at 300 dpi yields 7,200,000 pixels
3. 8.5x11" print at 300 dpi yields 8,415,000 pixels
4. 11x17" print at 300 dpi yields 16,800,000 pixels
5. 12x18" print at 300dpi yields 19,440,000 pixels
6. 13x19" print (A2 form) yields, 22,230,000 pixels
7. 16x20" print at 300dpi yields 28,800,000 pixels
8. 20x24" print at 300dpi yields 43,200,000, where at this resolution, the imager resolving power may exceed the resolving capabilities of the present day lenses; Canon's newer "L" lenses may perhaps handle this resolution.
My guess is that the next step after the 1DsMk2 (lets call it the Mark-III at the 2006 PMA show, would be the A2 page size at 22,230,000 pixels with a clean 25-3200 ISO range, which would shake-up the Medium format industry. I would imagine that 1DsMk2 will have a definite impact on the MF format industry. Imagine digital MF quality with a managible portable camera, with 60 lenses and huge accessories list to choose from in the example of the Canon line-up.
Jim Wilson. Professor
Florida Atlantic University
Boca Raton, FL
[email protected]
Truth is that CANON and SONY are capable of producing the 43MP imager shown below (today), however, it would only be affordable by NASA. The time and materials based progression of steps, is based on, lets say for example, Canon's ability to improve design, fabrication techniques, materials, and the spread of engineering costs over an increasing number of deliverable units.
Do the math, 11x17x300x300 = 16,800,000 approx. This is two 8.5x11" page centerfold, set by the USA print industry (medium format territory). The next pixel count steps are, from a "what are you going to do with the images" perspective:
1. 6"x8" print at 300 dpi yields 4,320,000 pixels
2. 8"x10" print at 300 dpi yields 7,200,000 pixels
3. 8.5x11" print at 300 dpi yields 8,415,000 pixels
4. 11x17" print at 300 dpi yields 16,800,000 pixels
5. 12x18" print at 300dpi yields 19,440,000 pixels
6. 13x19" print (A2 form) yields, 22,230,000 pixels
7. 16x20" print at 300dpi yields 28,800,000 pixels
8. 20x24" print at 300dpi yields 43,200,000, where at this resolution, the imager resolving power may exceed the resolving capabilities of the present day lenses; Canon's newer "L" lenses may perhaps handle this resolution.
My guess is that the next step after the 1DsMk2 (lets call it the Mark-III at the 2006 PMA show, would be the A2 page size at 22,230,000 pixels with a clean 25-3200 ISO range, which would shake-up the Medium format industry. I would imagine that 1DsMk2 will have a definite impact on the MF format industry. Imagine digital MF quality with a managible portable camera, with 60 lenses and huge accessories list to choose from in the example of the Canon line-up.
Jim Wilson. Professor
Florida Atlantic University
Boca Raton, FL
[email protected]
James L Wilson
Senior Member
--
James L Wilson, Digital Imaging Systems enterprise, http://www.BocaRatonDigital.com , Raton, Florida, Director and Principal Officer
George181996
Leading Member
Remember that, back in the late 80's, Apple were using a Windows environment, while Intel and Microsoft exploited first the x286 user base, before finally jumping into their Windows environment.
There are a lot of cases which document deliberate slowing down of the market in order to fully exploit an existing product. Remember also, the delay in the development of flat television screens, a case in which a lot of companies were involved.
Even now, I would not exclude the possibility of a silent agreement between Nikon and Canon (who control the DSLR market) in some points.
regards,
George
We were working with scanned images that were much larger than what a 25 meg sensor would output years ago and did fine with the computers then
Tom Verbeure
Member
I think you're 100% wrong.
They could have made some basic logic (an adder, a shifter, some memory cells) at this speed, but not a x86 compatible processor. The main reason is power, at this moment the whole industry is already struggling with this. At one time, you could overcome this easily by going to smaller transistor sizes, but these tricks don't work as easily anymore. (Due to transistor leakage is you'd like to know.)
Intel has an incredible research budget, but they don't control all the equipement that is deployed in their fabs. These are machines produced by Applied Materials, Canon, ... Before a new technology can be deployed (which would be necessary for 10 GHz processors), all pieces would have to fall in place before Intel can even think about starting to tune the processor. At this very moment, there are still major discussions going on between equipment manufacturers about specifications that are essential to make the machines together. Intel is undoubtely a major player in this, but they don't own these companies.
Intel is already stepping back from the MHz race and moving toward multi-processor chips that run at a lower speed. I'd very surprised to see a 10GHz processor before 2007.
If 10 years in the semiconductor industry has teached me something, it is that 3.6 GHz is an amazing feat. The vast majority of chip companies (say, 95%) creates chips that run at less than 400 Mhz. Above that, it becomes very difficult.
The statement above is basically total BS.
It could have been true at one point, say 10 years ago, when transistor switching speed was the main contributor to the overall speed of a processor, but that area is unfortunately long behind us.
The higher the amount of logic, the more area is needed to place it on the chip. The more area, the bigger the distance to interconnect the logic. At this moment, in 65nm technology, the switching speed of a transistor accounts for only 25% of the total delay. The rest is propagation delay to interconnect the logic. The only way to counteract this is to increase the amount of pipelining, as Intel has down for the P4 and P4E. These processors even have stages that do absolutely nothing: they are only inserted to move signals from one place to the other. The problem with pipelining is that it makes the power consumption go up dramatically AND it increases so-called misprediction penalties that can happen when the processor doesn't know which instruction to execute next. As a result, the processor becomes more inefficient. Throwing more logic at it is not going to solve the problem.
As transistor size goes down, the speed of the transistor still increases but interconnect delay increases too, mainly because resistance of the metal wires increases and because you get more capacitive coupling between wires because they are now closer to eachother. Since the ratio is already 25% 75% additional transistor speed is not going to help you much.
I'm sure that at one point or another solutions will be found, but the current state of technology isn't there yet. Intel can throw as much money as it wants at research, at one point it won't accelerate anymore an invention process that can take years to materialize (especially since there are so many other players involved.)
Moores Curves predicted a steady increase in transistor density. This trend is still more or less in place. But Moore did not make any predictions about a steadily increasing clock speed, even though that was initially also the case (as long as switching speed was the major contributing factor.)
In other words, your mail would have made somewhat sense 10 years ago, but right now it's totally out of reality.
Tom
Bo
Forum Enthusiast
It's all about AMD and Intel. It's the same with Canon and Nikon. They will always come up with the newest best thing to one-up the other. I think Canon has just gone a little farther this time with both a FF sensor and 17mp.
Bo
Bo
D
David Martin
Guest
Yes, Bernard, getting good performace at the edges needs a good lens.
But the difficulties do not stop there.
You indicate that Canon should be able to produce sensors for the 1Ds2 at 20D type resolutions at essentially similar costs, given the lower yields of FF.
This is surely not the case.
The greater angle of incidence due to the curvature of even the best lenses means that less light is received by the photosites at the edges of the sensor.
The larger size of the pixels in the 1ds2 vis-a-vis the 20D means that this is compensated by the larger pixel size, so that you get a similar amount of light on the edge pixels of the 1Ds2 to the pixels over the chip on the 20D.
Therefore essentially all you need to do is 'de-tune' or 'de-sensitise' the output from the cental photosites on the 1Ds2 to obtain similar results to the 20D.
This is a lot easier to do than increasing the performance of the edge pixels, which is what you would have to do to have the size of pixels throughout the 1Ds2 the same as in the 20D.
If you can manage that, you might as well reduce still further the pixel size in the 20D!
You have essentially had to crack the next step up in technology to do as you suggest.
So what it boiks down to is that Canon have not limited the resolution in the 1Ds2 for some obscure reason of their own, or even due to storage or other issues, sut simply because that is what is possible with the technology at the moment.
In order to demonstrate your point it would be neccessary to show:
First that you could obtain the same performance on the pixels at the edge at no extra cost in spite of their having less light to work with.
Second, if you denmonstrate the first, why in that case Canon did not do so, other than from some spirit of petty malice.
Regards,
DaveMart
Please see profile for equipment
But the difficulties do not stop there.
You indicate that Canon should be able to produce sensors for the 1Ds2 at 20D type resolutions at essentially similar costs, given the lower yields of FF.
This is surely not the case.
The greater angle of incidence due to the curvature of even the best lenses means that less light is received by the photosites at the edges of the sensor.
The larger size of the pixels in the 1ds2 vis-a-vis the 20D means that this is compensated by the larger pixel size, so that you get a similar amount of light on the edge pixels of the 1Ds2 to the pixels over the chip on the 20D.
Therefore essentially all you need to do is 'de-tune' or 'de-sensitise' the output from the cental photosites on the 1Ds2 to obtain similar results to the 20D.
This is a lot easier to do than increasing the performance of the edge pixels, which is what you would have to do to have the size of pixels throughout the 1Ds2 the same as in the 20D.
If you can manage that, you might as well reduce still further the pixel size in the 20D!
You have essentially had to crack the next step up in technology to do as you suggest.
So what it boiks down to is that Canon have not limited the resolution in the 1Ds2 for some obscure reason of their own, or even due to storage or other issues, sut simply because that is what is possible with the technology at the moment.
In order to demonstrate your point it would be neccessary to show:
First that you could obtain the same performance on the pixels at the edge at no extra cost in spite of their having less light to work with.
Second, if you denmonstrate the first, why in that case Canon did not do so, other than from some spirit of petty malice.
--
Regards,
DaveMart
Please see profile for equipment
Bernard Languillier
Senior Member
Dear David,
Interesting discussion.
Although you are right from a theoretical point of view, we have no way to know if Canon is close to the limit with the current 16.7 Mp sensor for the edge photosites. Can you proof that they cannot easily compensate this with more analog gain while still controlling the noise? I believe that you cannot, can you?
The answer to your second question is my very point since the beginning. It is possible that Canon didn't do it because they knew that people buying a 16.7 1DsMKII today would buy another 22MP D2 in 2 years from now which doubles the amount of money they get for a given level of performance released. This is not malice, and I never implied it was, this is business/marketing.
Best regards,
Bernard
Interesting discussion.
Although you are right from a theoretical point of view, we have no way to know if Canon is close to the limit with the current 16.7 Mp sensor for the edge photosites. Can you proof that they cannot easily compensate this with more analog gain while still controlling the noise? I believe that you cannot, can you?
The answer to your second question is my very point since the beginning. It is possible that Canon didn't do it because they knew that people buying a 16.7 1DsMKII today would buy another 22MP D2 in 2 years from now which doubles the amount of money they get for a given level of performance released. This is not malice, and I never implied it was, this is business/marketing.
Best regards,
Bernard
Tom Verbeure
Member
Just a question: why do you think the 20D sensor is manufactured in a different fab?
The pixel site size or the process does not mean that a different fab needs to be used. It's just a matter of changing the parameters of the fab machinery, which can be changed easily after each batch of wafers.
E.g. the fab at my previous company was able to produce anything from 1um downto 0.35um technology designs. Some chips could have extra processing steps to add analog features. Some have high voltage capabilities others don't. Etc etc.
Tom
D
David Martin
Guest
Well, Bernard, it appears to me that any onus of proof would be upon you, as the current design for the 1Ds2 does not perform as you say it*should*, and what's more you have indicated that it should be no more expensive.
Since as you now say there are cogent theoretical reasons for inferring it is in fact much more difficult to produce a FF sensor at the same pixel pitch as the 20D, your statement that they could do so seems entirely without foundation.
They could, however, if they wished assuredly re-engineer the 1D2 to have 12MP as edge issues would not be significant.
In oredr to keep costs to the same levelas the 1D2, it would however be neccesary to reduce the frame rate to 6fps to have the same data throughput.
That they did not do so is due firstly to the less suitable nature of 6fps for the intended sports use, and is also an indication of how fast Canon are moving in improving the performance of their sensors.
Possible, but there is no evidence at all that it is in fact the case.
The 1Ds2 improves upon the pixel count on the 1Ds very substantially already.
--
Regards,
DaveMart
Please see profile for equipment
D
David Martin
Guest
Alan, I agree the 1Ds does not display difficulties due to edge issues.
But designing a sensor to dope with the curvature in even the best lenses at the edges when they are not over-sized, as FF lenses are on APS -sized sensors, is tough and Canon had to go for a more conservative pixel pitch to do this - thas why the pixels ont eh 1Ds2 are bigger than on the 20D., just as the pixels on the 1Ds were bigger than those on the D60.
But with that advantage in hand Canon did a great job of it, and the !Ds, and likely the 1Ds2 do not display any difficluties to the user due to edge issue.
Good job Canon!
Regards,
DaveMart
Please see profile for equipment
But designing a sensor to dope with the curvature in even the best lenses at the edges when they are not over-sized, as FF lenses are on APS -sized sensors, is tough and Canon had to go for a more conservative pixel pitch to do this - thas why the pixels ont eh 1Ds2 are bigger than on the 20D., just as the pixels on the 1Ds were bigger than those on the D60.
But with that advantage in hand Canon did a great job of it, and the !Ds, and likely the 1Ds2 do not display any difficluties to the user due to edge issue.
Good job Canon!
--
Regards,
DaveMart
Please see profile for equipment
i would just like to say two things:
first, EOS 1Ds is top of the line camera. it represents best that canon can manufacture, and by some chance is probably most advanced digital camera on the market. it's a professional no-compromise camera, and since canon is very much into R&D (as far as CMOS sensors go), perhaps they simply felt that this 16,7MP chip would offer best ration of performance and resolution. even if they had higher resolution chip, they may have chosen not to use it because it still had problems (be it more noise, higher power consumtion, etc.) that would probably be resolved.
second thing, maybe they wanted to maintain reasonable amount of FPS, which with limited bus throughoutput means limited resolution. maybe even digic II couldn't handle enough data per second?
maybe the whole camera wasn't ready? maybe lens lack optical resolution? maybe their manufacturing lines couldn't handle such sensor density and produce much more faulty sensors, which would all be too expensive?
...all that given, of course, that they had better CMOS ready for production.
finally, if they had 22MP sensor ready and none of the above problems, why not built it in the camera now, knowing that with technological advancement they'll have 30MP sensor in 2-3 years?
so all in all, i doubt this is the case.
intel and AMD have different kind of problems, and that's manufacturing transistors on superatomic levels: they reached the point where one transitor consists of several dozen atoms. it's hard to use the same manufacturing process (and same set of physics laws ;-) and push this technology much further. so they lag deliberately... waiting for deus-ex-machina solution, like quantum computing technology, trying to make some money in the process.
greets,
matija
--
photo: blushark.deviantart.com/gallery/
design: bushidocat.deviantart.com/gallery/
first, EOS 1Ds is top of the line camera. it represents best that canon can manufacture, and by some chance is probably most advanced digital camera on the market. it's a professional no-compromise camera, and since canon is very much into R&D (as far as CMOS sensors go), perhaps they simply felt that this 16,7MP chip would offer best ration of performance and resolution. even if they had higher resolution chip, they may have chosen not to use it because it still had problems (be it more noise, higher power consumtion, etc.) that would probably be resolved.
second thing, maybe they wanted to maintain reasonable amount of FPS, which with limited bus throughoutput means limited resolution. maybe even digic II couldn't handle enough data per second?
maybe the whole camera wasn't ready? maybe lens lack optical resolution? maybe their manufacturing lines couldn't handle such sensor density and produce much more faulty sensors, which would all be too expensive?
...all that given, of course, that they had better CMOS ready for production.
finally, if they had 22MP sensor ready and none of the above problems, why not built it in the camera now, knowing that with technological advancement they'll have 30MP sensor in 2-3 years?
so all in all, i doubt this is the case.
intel and AMD have different kind of problems, and that's manufacturing transistors on superatomic levels: they reached the point where one transitor consists of several dozen atoms. it's hard to use the same manufacturing process (and same set of physics laws ;-) and push this technology much further. so they lag deliberately... waiting for deus-ex-machina solution, like quantum computing technology, trying to make some money in the process.
greets,
matija
--
photo: blushark.deviantart.com/gallery/
design: bushidocat.deviantart.com/gallery/
David J. Littleboy
Senior Member
James L Wilson wrote:
Quite sensible things. Yep, 16.7 MP is a lovely pixel count. Enough pixels to print at 11x17 (A3) at 300 dpi and at 16x24 at 200 dpi. (If your audience isn't putting their noses on your prints (as they tend not to with larger prints on gallery walls) 200 dpi from dSLR images looks very good).
More pixels means more noise, and the main problem with the 1Ds was that it's noise levels were a lot worse than the other Canon dSLRs. Canon needed to bring their top-of-the-line model up to the standards of the rest of the line.
And higher pixel counts run into severe diminishing returns because lens resolution can't keep up.
16.7MP is simply perfect.
Minor quibbles follow.
A2 is 16.5 x 23.
David J. Littleboy
Tokyo, Japan
Quite sensible things. Yep, 16.7 MP is a lovely pixel count. Enough pixels to print at 11x17 (A3) at 300 dpi and at 16x24 at 200 dpi. (If your audience isn't putting their noses on your prints (as they tend not to with larger prints on gallery walls) 200 dpi from dSLR images looks very good).
More pixels means more noise, and the main problem with the 1Ds was that it's noise levels were a lot worse than the other Canon dSLRs. Canon needed to bring their top-of-the-line model up to the standards of the rest of the line.
And higher pixel counts run into severe diminishing returns because lens resolution can't keep up.
16.7MP is simply perfect.
Minor quibbles follow.
This is also "A4", 8.25 x 11.5
This is also "A3", 11.5 x 16.5.
This is "A3 nobi" in Japanese, Super A3 to the rest of us, a size not in the ISO standard.
A2 is 16.5 x 23.
The 1Ds2 will shake up the MF industry. It completely obviates the need for 645 and 6x6, and 6x7 will be so little better as to be not worth the bother.
Well, except for the problem that most MF cameras are a lot lighter and better handling than the 1DsM2. I used to call my Mamiya 645 the Jumbo Martian Death Ray until I hefted a 1Ds in a store...
David J. Littleboy
Tokyo, Japan
Keyboard shortcuts
- f
- Forum