Why not a low megapixel full frame?

[lucas_jarvis]

From what I understand, the more you crowd the pixels, the more

noise you will get. Why not make a full frame sensor with around 8

megapixels? Maybe you would lose a very slight edge in resolution,

but it would surely have a very useable 3200 iso. But maybe it

wouldn't lose in resolution, after all you would be able to use

faster shutter speeds and smaller apertures. (Maybe I'm stretching

it a bit) If you're not likely to go over 11x14 I think this would

be a very desireable camera. Heck, I don't think if you printed

large it would be an issue anyway. Anyone who makes large prints

knows how to properly upsize in photoshop. Is it just the mentality

of comsumers caught in the megapixel race that would prevent this

scenario from happening, or have I overlooked something?

[bobatkins]

I don't think it's particularly cheaper to make a lower pixel count sensor once you've developed the technology to make a higher pixel count. You might get a slightly higher yield, but I doubt that it's costing Canon much more to make their 8MP sensors than it did their 6MP sensors.

 

Noise might be slightly lower at high ISOs, but most people don't shoot at high ISOs. Don't foget the 12MP FF sensors have larger pixels than the 8MP APS-C sensors do and they aren't really much less noisy.

 

Since Canon can obviously make 12MP full frame sensors that work pretty well, I doubt it would pay them to make a FF 8MP sensor.

[carlos_miami]

It probably wouldn't have a market large enough to justify the development costs.

 

What I want to know is, why is my 16MP digital back with a 36x36mm sensor so awful at ISO400? Is it because its software designers figured that it would not be used at high ISO and optimized it for low ISO use?

[lovcom_photo]

I would rather have wider Dynamic range then more pixels, personally.

 

I would sell off my 12.8mp 5D to get a 8-10mp camera that gave me 2-3 more stops of DR, and I'd even pay $1,000 more for it then what the 5D costs.

[etan_lightstone]

Lucas, more than likely it isn't a technology issue.. its a marketing issue.

 

The last thing canon or nikon wants to do is threaten sales of their high end cameras (or low end cameras for that matter) by introducing products that blur the ligns between their consumer cameras and their pro cameras.

 

An affordable 8mp version of the 5D or 1ds would be a serious threat.

 

And might I add that I've made perfectly sharp prints of 12x18 in size with my 6 megapixel Canon 300D.

[andrew robertson]

"I would rather have wider Dynamic range then more pixels, personally.

I would sell off my 12.8mp 5D to get a 8-10mp camera that gave me 2-3 more stops of DR,

and I'd even pay $1,000 more for it then what the 5D costs."

 

How would you use it? Output materials have a narrower dynamic range than even slide film

does.

[peter_white2]

Andrew,

 

Could you expand on that please? How are printers and papers rated for dynamic range? And how do they compress DR, if that's what you're saying?

 

Thanks,

[andrew robertson]

They don't really compress it - they leave out what they can't reproduce. Individual paper +

printer + ink combinations can be profiled, and your profile will tell the output device how

to deal with out of gamut information (i.e. it will make the stuff that is too dark brighter,

and the stuff that is too bright dimmer).

[fourfa]

"How would you use it? Output materials have a narrower dynamic range than even slide film does."

 

Same way people used negative films that have greater dynamic range than paper, wouldn't you guess? Nothing new here.

[savas_kyprianides]

Speaking of which, people talk about taking two shots and blending them in order to create

more range. Would both ends of a bracketed shot do for this purpose, omitting the middle

one?

[savas_kyprianides]

Speaking of which, people talk about taking two shots and blending them in order to create

more range. Would both ends of a bracketed shot do for this purpose, omitting the middle

one?

[mark u]

If your concern is about cost, then you just need a little patience. We are now in the phase when full frame sensors are dropping rapidly in cost (this is no longer true of APS sensors, where the cost drop has already happened), as yields increase. Full frame sensors already cost less than 1/5th of what they did when the 1Ds was launched, and as the technology for making them matures they can be expected to drop by a further factor of 2-3. As Bob says, the difference in cost is not really related to megapixel count - it's mostly about the frequency of impurities in the silicon wafers and the ability to use larger wafers that lead to less wastage.

 

If your concern is about dynamic range and high ISO, then larger pixels can certainly help here. That too is probably coming in the form of photosites that stack the RGB pixels above each other. There is already the Foveon technology that does this using the fact that different wavelengths of light are typically absorbed at different depths in silicon. A new Foveon sensor is expected this year which may solve some of the limitations of the one used in Sigma's SD10 camera - although it probably will only be in a cropped size. One of the things that the Foveon sensors have shown is that the effective resolution compared with a Bayer sensor with the same number of photosites is about double, because interpolation of the Bayer pattern doesn't have to be performed.

 

In addition to Foveon's technology, Fuji has just had a patent published that achieves the stacking by using pigments to enhance the colour sensitivity of the layered pixels (although the technology is still 3-4 years from being commercialised):

 

http://www.steves-digicams.com/diginews.html#fuji_sensor

 

http://www.nhk.or.jp/strl/open2005/en/tenji/t03.html

 

It has long been rumoured that Canon are working on something similar - the more recent rumours suggest it might see the light of day next year.

 

Among the advantages of photosites with stacked pixels is that you don't need any anti-alias filter to spread the light to adjoining pixels in order to get colour sampling at all points of the image (although you might use a weak AA filter to reduce Moire - but Moire is much less troublesome when you have stacked pixels because it doesn't produce the garish colour that comes with Moire on Bayer sensors - its misrepresentation of high frequency detail is much more graceful). The reduced need for AA filtering (or possibility of disensing with it) has other advantages in that there is less interaction with sensor optics to cause problems such as purple fringeing. Another advantage is that potentially you aren't throwing away 2/3rds of the light because each Bayer pixel is only sampling 1/3rd of the spectrum.

 

The difficulties are with ensuring adequate separation of the colour channels due to crosstalk (electrons leaking from one channel to the one next to it or photons being absorbed in the "wrong" channel), ensuring low noise (you can't as easily apply a voltage to drain the sensor of spurious electrons prior to exposure), and getting a good well depth for each channel to maximise dynamic range.

[John Crowe]

As Spock would say, with upturned brow..."fascinating".

[mark_man]

Thanks for the useful info Mark. Great to see DSLR sensor technology has some life in it yet - would be soul-destroying if we were stuck with what's currently available...

[rdkirk]

From what little Canon has said about their processes, it appears that they are capitalizing more on overall manufacturing efficiency than on actual decreasing sensor costs. It's amazing that with each model--not just each generation--they've improved design and assembly to the extent that it's cheaper to create a new sensor than to re-use an old one.

 

Notice the 350XT has it's own sensor even though the output is hardly different from the 20D. Although the 30D has changed little, it also has a different sensor--they apparently aren't continuing to use the old 20D line, not because of the sensor itself but because of streamlined design and assembly of the entire product.

 

I'm not sure real dynamic range is a problem with the current generation of sensors. Every comparison with film I've seen lately has been a false one...such as comparing negative film's tolerance for overexposure with a digital's sensor lack of tolerance for the same (without recognizing that a digital sensor has commensurably greater tolerance for underexposure than negative film).

 

When you compare paper to paper, I haven't seen any less dynamic range in prints made from modern digital equipment than from prints made from film...given equal expertise in the respective printing processes. It's amazing what you can do with a properly exposed RAW image and Fred Miranda's "Shadow Recovery."

[ilkka_nissila]

The dynamic range of paper or printers does not matter here. With digital printing, you can map any tone into any other tone. Thus you can start with a high contrast scene, photograph it with a wide dynamic range camera and print it as a low contrast photo on inkjet paper. This was done conventionally using type C printing and there is nothing difficult about it.

[andrew robertson]

Yes, of course that's true. However, you can only create the illusion of broad tonal range on

paper. In reality it is quite narrower than an average DSLR's sensor.

[kelly_flanigan1]

A larger surface area sensor costs more to produce, than a smaller one., one gets less sensors per wafer. This fact has been known in the wafer/IC/now sensor industry. Not understanding this might be difficult for dreamers. :) Yes I want a cost cost FF camera too! Imagine going to Home Depot and discovering that 5 lbs of nails cost more than 1lbs of nails. :) A high tech small stainless nail might cost the same per nail as an imported larger Common framing nail. In either case there is a base material cost of the steel type before making the nail. For a given material (nail, paper, sensor, optical glass), usually there is tend of a larger physical volume of the item, they tend to cost more. <BR><BR>With items made in super super high volumes, like Roller skate ball bearings, one gets a Abec 5, 7 or 9 precision class bearing radically lower in cost than a low production bearing.<BR><BR>Unless there is alot of production volume, full frame sensors are fighting the basic tooling and debugging costs to spread out over the small questionable production volume. With a high volume P&S camera's digital sensor, these costs get spread out across ALOT of production volume. <BR><BR>The catch 22 with "full frame" is that costs wont drop until volume increases, and some folks with their cache of Nikkors and Canon lenses wont buy until prices get reasonable.

[ilkka_nissila]

Well, any full frame DSLR from Nikon would be a start, whatever the price. They won't get money from selling them before they try.

 

I would be willing to pay about $6000 for a 1.0-1.2x DSLR from Nikon with otherwise D200-like specs (can be slower than 5 fps)). I would not be able to buy one immediately but would do so eventually.

[mark u]

Mr. Flannigan's math may give the impression that manufacturing improvements through experience depend on the number of sensors produced, irrespective of their size. In practice, I would suggest that the number of wafers used is a better guide. Since a 1/2.5" sensor has an area of just 24.7 sq mm, compared with 864 sq mm for a full frame one, it means that at least 35 times as many tiny sensors can be produced from one wafer (in fact it is rather more, because there is less wastage close to the edges of the circular wafer). For those tiny sensors (cost now similar to no more than a roll of film and D&P), yields are now well over 90% - possibly even up to 98% - compared with current full frame yields that are known to be over 25%. Being generous, that means a further factor of 3, or in total 3x35 ~=100 times as many viable sensors per wafer. Therefore, full frame sensors only need to capture 1% of the market for 1/2.5" ones for the rate of manufacturing experience accumulation in terms of wafers produced to be similar. Canon's digicam sales are ~20 million units per year - so that 1% is just 200,000 cameras, or about 10% of Canon's DSLR sales.

 

It's fairly obvious that Canon is market testing the price/volume relationship for the 5D currently, with pricing down to a US equivalent of $2,300 in Europe (i.e. before sales tax, post rebate). I'd guess they are hoping to be able to build sales volumes (perhaps after a further price reduction as manufacturing costs are lowered and sensor yields increased) to 25-30,000 cameras/month - plenty to keep the virtuous circle growing ahead of the next lower priced full frame body.

[ron c sunshine coast,qld,a]

The comments above about wider dynamic range being pointless is unbelievably dumb.Please think before posting

[kelly_flanigan1]

Another problem is that consumers like BIG NUMBERS. Thus a 6 or 8 megapixel full frame 35mm might have trouble against a 8 or 10 megapixel P&S digital, that costs 1/3 to 1/4 the price. Gurus would perfer the Full frame beast, but the soccer moms would flock to the "better camera with bigger numbers" . Like it or not folks really brag about the megapixel size of their cameras, and one that is just moderate doesnt bode well with "keeping up with the Joneses". A full frame dslr of moderate resolution would have to have good marketing, to not get lost in the <b>megapixel race</b>. I would perfer a rig like this because I like alot of low light work.

[mark u]

The megapixel wars are now over. The ISO wars have begun.

[lucas_jarvis]

I welcome the ISO wars. Could silky smooth iso 6400 be just around the corner!? :)

[kelly_flanigan1]

It seems every few months the highset megapixel camera at Frys, Beet Buy, Walmart or Walgreens is higher. The megapixel wars are still going in P&S digitals. Walmart has dinky 8 megapixel cameras, Walgreens 3 megapixel stater cameras. Many of the midwest Walmarts that use to carry a Canon Drebel or its newer replacement dont carry them anymore. With local camera stores dying off, will dslrs be a speciality item only available by mailorder?