Canon Announces the Development of New High Sensitivity Sensor

[unfocused]

Seriously, EVERY review I've read and every comparison I've looked at makes it clear that Canon's lineup of sensors far outperforms Nikon's in high ISO performance.

???

The 1DX is probably only barely better at high ISO than the D4 and it also, I think, has a weaker CFA.
Canon isn't worse at high ISO now but far better???

I wouldn't say the Canon is "far" better...things are limited largely by physics at that point. In real-world examples, I've noticed more color noise from the D4 at ISO 25600 and 51200 (probably because those settings are digitally amplified, vs. Canon's primarily analog amplification). Outside of that very slight difference, the two cameras are definitely comparable at those levels...you would be hard pressed to notice any real differences in most situations, I think.

Okay. Point well-taken. Post corrected. Strike "far." And, frankly, I may have overreached a bit referring to the 1DX. Upon further consideration, I would say that most of the comparisons I have seen have been between the 5DIII and the D800. General consensus is that the Nikon is great for resolution, but falls apart at higher ISOs, while the 5dIII shines at high ISOs. Early reviews seem to indicate the 6D also performs quite well at high ISOs and from the comparisons I've seen between the 7D and the 24mp Nikons, the Canon 18mp sensor still holds its own against the competition despite its age.

Main point still stands. There is no "legendary" low-light performance on Nikon sensors.

 

[jrista]

Wow. This might go head-to-head (if not surpass) Nikon's current lineup of legendary low-light performance sensors.

Maybe. There are other things that can add noise besides dark current. Sony Exmor converts sensor data to a digital form as early on in the pipeline as they can, and they gain some benefits from that. They also parallelize processing, allowing lower frequency components, as well as isolating high frequency components away from pixel processors. That minimizes the components in the pipeline that can add additional noise, and minimizes the length of analog components in the pipeline.

So long as Canon continues to use off-die, high frequency ADC and any down-stream amplifiers, I don't foresee their IQ at low ISO being much better than it is now on current Canon sensors, and not as good as Exmor. Dark current is only one of a few primary contributors to noise...the whole pipeline needs to be addressed to compete or surpass the Sony/Nikon alliance and Exmor.

 

[jrista]

Main point still stands. There is no "legendary" low-light performance on Nikon sensors.

Well, in that case, neither are there any "legendary" low-light performance Canon sensors. Canon has "the edge" compared to the D4, but things are really physics-bound at really high ISO, and the margin for improvement is limited. There is a lot more room to grow and "be legendary" on the low ISO end of things, as Canon is currently performing about 2 stops less than the 14-bit depth of their current cameras theoretically allow. A move to 16-bit sensors, assuming Canon still can only produce ~12 stops of DR, would mean they are FOUR STOPS behind the curve.

I think it's tough to really be legendary at ultra high ISO. Things are just too limited, especially for bayer-type sensors. A move to monochrome, or to a color-splitting rather than color-filtering approach, might increase the headroom there and allow a moderately significant jump forward (imagine ISO 204800 as good as ISO 12800-25600 on the 1D X now). I don't foresee color splitting in Canon or Nikon cameras any time particularly soon, and increasing pixel area, along with reducing sensor temperature, are probably the two most effective ways right now to improve photodiode Q.E. and improve high ISO performance...but the gains won't be massive.

 

[AvTvM]

Can anyone explain why DSLRs sensors are not square which would provide more viewing area?

This has been gone over more times on this forum than I would ever count. Simple answer: a square sensor with a diagonal of 43.3mm (same as 36x24mm) would not work due to the extra height needed for the reflex mirror and the flange distance of the EOS system (the taller mirror would hit the rear element or mount of the lens). And as others have pointed out, not all lenses have round baffles to produce the full image circle (only to cover the portion of the image circle that contains the 36x24mm frame.

simple solution. a mirrorless camera! Bingo!
Which happens to be exactly what I want.
Even if it does not have a 36x36mm sensor. Although I would gladly take one.

 

[pedro]

Main point still stands. There is no "legendary" low-light performance on Nikon sensors.

I think it's tough to really be legendary at ultra high ISO. Things are just too limited, especially for bayer-type sensors. A move to monochrome, or to a color-splitting rather than color-filtering approach, might increase the headroom there and allow a moderately significant jump forward (imagine ISO 204800 as good as ISO 12800-25600 on the 1D X now). I don't foresee color splitting in Canon or Nikon cameras any time particularly soon, and increasing pixel area, along with reducing sensor temperature, are probably the two most effective ways right now to improve photodiode Q.E. and improve high ISO performance...but the gains won't be massive.

@jrista: "Imagine ISO 204800 as good as ISO 12800-25600 on the 1D X now"...Hope we get there in about 10 years from now...Likely?

 

[jrista]

Main point still stands. There is no "legendary" low-light performance on Nikon sensors.

I think it's tough to really be legendary at ultra high ISO. Things are just too limited, especially for bayer-type sensors. A move to monochrome, or to a color-splitting rather than color-filtering approach, might increase the headroom there and allow a moderately significant jump forward (imagine ISO 204800 as good as ISO 12800-25600 on the 1D X now). I don't foresee color splitting in Canon or Nikon cameras any time particularly soon, and increasing pixel area, along with reducing sensor temperature, are probably the two most effective ways right now to improve photodiode Q.E. and improve high ISO performance...but the gains won't be massive.

@jrista: "Imagine ISO 204800 as good as ISO 12800-25600 on the 1D X now"...Hope we get there in about 10 years from now...Likely?

Who knows, really. If manufacturers stick to 3-4 year release schedules, it might indeed take that long. Technologically, I think we could probably achieve that today (assuming companies like Canon, Nikon, and Sony had infinite R&D budgets. )

 

[RGF]

So if I get this right, Canon is taking a full frame sensor, and ENLARGING the pixels on it 7.5 times, and then market it as the best ISO performer on the market? DUH!

So the pixel count will be 18MP / 7.5^2 = 0,3 MP on a FF?

Ehm, I might have completely misunderstood, but how is this technical progress?

actually divide by the difference in area, not area squared. 18 mp / 7.5 = 2+ MP on full frame.

 

[qwerty]

Main point still stands. There is no "legendary" low-light performance on Nikon sensors.

I think it's tough to really be legendary at ultra high ISO. Things are just too limited, especially for bayer-type sensors. A move to monochrome, or to a color-splitting rather than color-filtering approach, might increase the headroom there and allow a moderately significant jump forward (imagine ISO 204800 as good as ISO 12800-25600 on the 1D X now). I don't foresee color splitting in Canon or Nikon cameras any time particularly soon, and increasing pixel area, along with reducing sensor temperature, are probably the two most effective ways right now to improve photodiode Q.E. and improve high ISO performance...but the gains won't be massive.

@jrista: "Imagine ISO 204800 as good as ISO 12800-25600 on the 1D X now"...Hope we get there in about 10 years from now...Likely?

Who knows, really. If manufacturers stick to 3-4 year release schedules, it might indeed take that long. Technologically, I think we could probably achieve that today (assuming companies like Canon, Nikon, and Sony had infinite R&D budgets. )

You might get your ISO 204800 thats as good as ISO 12800-25600, but its not going to come from sensor efficiency improvements. You are asking for a 3-4 stop improvement. Using jrista's numbers earlier in this thread, going from capturing 17.5% of photons to 45% just gives you 1.36 stops (log(45/17.5)/log(2))... even going to 100% photon capture efficiency would just give you 2.5 stops.

I see numbers for quoted QEs for modern sensors at about 50% (see http://sensorgen.info/), which means that you can only get one more stop out of them (presumably these efficiencies are measured behind the color filter; as jrista points out and I analyze above, you can get a bit more than one by bypassing the color filter)

You might be able to get some benefit from active cooling, but it won't help with photon shot noise. Anyone have the calcs handy for how many stops (or what fraction of a stop) cooling your sensor with liquid helium (and magically avoiding condensation would buy you)?

The big hope is of course for better image processing algorithms that make the final images "look" less noisy (even though the raw data they are derived from has the same amount of noise). The upshot to this is your raw images you take today will look better too.

Of course, bigger sensors help too. I will buy an 8x10 format mirrorless with per-pixel specs matching the current DSLRs just as soon as I can afford it. It won't even need active cooling, what with hell being frozen over and all.

 

[jrista]

Main point still stands. There is no "legendary" low-light performance on Nikon sensors.

I think it's tough to really be legendary at ultra high ISO. Things are just too limited, especially for bayer-type sensors. A move to monochrome, or to a color-splitting rather than color-filtering approach, might increase the headroom there and allow a moderately significant jump forward (imagine ISO 204800 as good as ISO 12800-25600 on the 1D X now). I don't foresee color splitting in Canon or Nikon cameras any time particularly soon, and increasing pixel area, along with reducing sensor temperature, are probably the two most effective ways right now to improve photodiode Q.E. and improve high ISO performance...but the gains won't be massive.

@jrista: "Imagine ISO 204800 as good as ISO 12800-25600 on the 1D X now"...Hope we get there in about 10 years from now...Likely?

Who knows, really. If manufacturers stick to 3-4 year release schedules, it might indeed take that long. Technologically, I think we could probably achieve that today (assuming companies like Canon, Nikon, and Sony had infinite R&D budgets. )

You might get your ISO 204800 thats as good as ISO 12800-25600, but its not going to come from sensor efficiency improvements. You are asking for a 3-4 stop improvement. Using jrista's numbers earlier in this thread, going from capturing 17.5% of photons to 45% just gives you 1.36 stops (log(45/17.5)/log(2))... even going to 100% photon capture efficiency would just give you 2.5 stops.

I see numbers for quoted QEs for modern sensors at about 50% (see http://sensorgen.info/), which means that you can only get one more stop out of them (presumably these efficiencies are measured behind the color filter; as jrista points out and I analyze above, you can get a bit more than one by bypassing the color filter)

You might be able to get some benefit from active cooling, but it won't help with photon shot noise. Anyone have the calcs handy for how many stops (or what fraction of a stop) cooling your sensor with liquid helium (and magically avoiding condensation would buy you)?

The big hope is of course for better image processing algorithms that make the final images "look" less noisy (even though the raw data they are derived from has the same amount of noise). The upshot to this is your raw images you take today will look better too.

Of course, bigger sensors help too. I will buy an 8x10 format mirrorless with per-pixel specs matching the current DSLRs just as soon as I can afford it. It won't even need active cooling, what with hell being frozen over and all.

That would be 2.5 stops for full-sensor sensitivity for current-sized sensors. Per-pixel sensitivity can increase by increasing area, so once you hit the limits of Q.E., you can continue improving per-pixel performance by making them larger. That obviously has the impact of reducing image resolution. You can continue getting better overall sensor performance at current pixel counts with larger sensors. Given how cheap FF is becoming, for low-light fiends, it might become practical to produce a larger sensor for the really high-end ($7000+) DSLR's that maintain pixel count with much larger pixels.

Of course...you need a new lens system to support that unless Canon can figure out a way to make the EF mount compatible...maybe with an adapter for current lenses like they did with EOS-M...

 

[Lee Jay]

Large pixels generally have high read-noise which means they tend to have a tough time in such environments. Read the text. They claim to have developed technologies that counteract that affect (that read noise increases with pixel size).

Hmm, ok. But in the general discussion, noise increases when pixels get smaller (at Hi-ISO).

Yeah, but that's a myth.

 

[tnargs]

I remember a canon experimental camera from a while ago, a weird white spaceship looking thing, talking about the future of photography being the imager videoing the subject then selecting the best frames. This is NOT a direction I want things going;

That reminds me a bit of the Canon N that came out last year.

All changes that happen in all areas happen despite the Old School moaning and fighting and withholding their credit cards at every step.

I am also reminded when Nintendo brought out the Wii. The Old School of video gamers pilloried and blasted and canned it from wall to wall and announced that this is the end of Nintendo because no one will buy it. They wanted a standard machine but with more video resolution, more speed, more audio. Just keep the same old formula but ever better.

Is there a parallel in photography, I wonder.

 

[Bosman]

I don't care about this. I think it was around 2006 or 2007 I heard about Canon having developed a 50mp sensor. Yippie 6/7 years later and still nothing from that that helps unless the 50mp sensor had the new Micro-lenses design that added to the sensors today. Certainly Canon hasn't advanced sensor tech enough to see major advancement other than high iso which is probably more a software than a hardware thing.
Ok i will be honest here about advancements. I realize that a 50mp sensor is of no use without major advancements in memory tech and decreases in memory storage costs. I still wouldn't want a 50mp camera 5 yrs from now for that reason. So i'll give them that but like the Nikon Mantra, wheres the Dynamic Range advancements? The fact that that tech is avail on Nikon means Canon users are without. I will also say have i missed it on my 5dm3, no, would i be happy to have increased DR, well of course i would? It would be nice to shoot and be several stops over and not blow out and kill color info. That to me would be advancement.
I do really like the whole shoot in the dark thing and i do see Canon start to market that more even like with the 6D so maybe there is something there. Basically the closer a camera can get to what an eye sees we are all better off.
Ahhh, I am glad i ended that on a pleasant note lol.

 

[qwerty]

That would be 2.5 stops for full-sensor sensitivity for current-sized sensors. Per-pixel sensitivity can increase by increasing area, so once you hit the limits of Q.E., you can continue improving per-pixel performance by making them larger. That obviously has the impact of reducing image resolution. You can continue getting better overall sensor performance at current pixel counts with larger sensors. Given how cheap FF is becoming, for low-light fiends, it might become practical to produce a larger sensor for the really high-end ($7000+) DSLR's that maintain pixel count with much larger pixels.

Of course...you need a new lens system to support that unless Canon can figure out a way to make the EF mount compatible...maybe with an adapter for current lenses like they did with EOS-M...

Here's hoping you're right about larger format sensors being the wave of the future. I would just as soon have the pixels stay the same size (i.e. more pixels overall), so that I can decide in post whether I want higher resolution or less (per-output-pixel) noise, instead of having the latter forced on me by default.

The problem is that its hard to lay out larger sensors on a silicon wafer, and that yield plummets as you make bigger chips. How many fully functional 645-size sensors (either true 645 or the smaller 645d digital size) do you think you can get out of a 300mm wafer, or even a 450mm one? (I did the layout for that once, and looked up some numbers for expected yield and per-wafer cost, and it was not encouraging. However, I am definitely not an EE or a fab specialist, so someone else might have a better idea.)

The nice thing about making a larger format mirrorless camera is that it would be trivial to make an adapter that would allow the use of existing EF and EF-S lenses (no worries about clearance for the larger mirror) with the obvious caveat that the image circle would not cover the entire sensor (and the less obvious caveat that I have never used a mirrorless camera with hybrid af and the contrast detection af I have used is relatively slow). Of course, the question is whether Canon would pass up the opportunity to sell everyone a new set of lenses.

 

[jrista]

Large pixels generally have high read-noise which means they tend to have a tough time in such environments. Read the text. They claim to have developed technologies that counteract that affect (that read noise increases with pixel size).

Hmm, ok. But in the general discussion, noise increases when pixels get smaller (at Hi-ISO).

Yeah, but that's a myth.

It is only a "myth" assuming images are always compared on a size-normal basis. That is certainly a valid way to compare, and the only normalized way to compare. However...assuming one buys a higher resolution camera for the purposes of using it for its native resolution, rather than downscaled to something smaller...the increased noise of a higher density sensor is no myth.

Smaller pixels have a lower cap on charge. Lower charge means higher gain. Higher gain means that for any given illumination level photon shot noise is exacerbated by amplification, which results in higher noise at native size.

 

[qwerty]

Large pixels generally have high read-noise which means they tend to have a tough time in such environments. Read the text. They claim to have developed technologies that counteract that affect (that read noise increases with pixel size).

Hmm, ok. But in the general discussion, noise increases when pixels get smaller (at Hi-ISO).

Yeah, but that's a myth.

It is only a "myth" assuming images are always compared on a size-normal basis. That is certainly a valid way to compare, and the only normalized way to compare. However...assuming one buys a higher resolution camera for the purposes of using it for its native resolution, rather than downscaled to something smaller...the increased noise of a higher density sensor is no myth.

Smaller pixels have a lower cap on charge. Lower charge means higher gain. Higher gain means that for any given illumination level photon shot noise is exacerbated by amplification, which results in higher noise at native size.

Normalized is the way to compare that makes sense for... well... pretty much anyone. Saying that you should compare images at native resolution because one buys the camera to shoot at native resolution does not make much sense.

Sure, a shot from a hypothetical 40MP camera would produce a worse 36x24" print at higher ISOs (if you looked at it up close) than a 18x12" print from a 10MP camera with the same sensor tech. However, the 40MP print and the 10MP print would look almost indistinguishable if you printed them both at 18x12". However, and this is the big one, if you printed them both at 36x24", the 40MP print would look better at low ISOs (and basically no worse under other conditions).

(Sidenote: I was going to use 4x6" and 8x12" as example print sizes, but both sensors have plenty of resolution for that print size. 24x36 should be just about the smallest size where 10MP isn't quite enough)

You don't buy a camera just to shoot at native resolution (at least, I don't); you buy a camera to take pictures (or make prints). For example, I was doing relatively low light indoor sports shots last weekend with a 5d iii (typically f/2-2.8, iso 6.4-12.8k, 1/250-1/500). I had to downsize the images to about 3-5MP before I was happy with the quality (in other words, at higher resolutions I was either seeing noise or blur from noise reduction). So I could have gotten the same printable results with a 5MP camera in that particular case. However, the 20-whatever MP of the 5diii let me make that decision after I took the picture. And, if the light had been better, or if I cared enough to really buckle down with the post processing and noise reduction, the extra MP would have given me the ability to get higher resolution images.

Thats not to say that there aren't advantages to lower resolution cameras - a 5MP 5diii would not fill up the buffer nearly as fast, and previews would load faster in lightroom (the storage space isn't really an issue for me). A 10MP camera makes decent enough prints at 24x36 (certainly to the point where my skill, rather than the equipment, is the biggest limiting factor). And I am sure you could design a different sensor for every resolution and get slightly better print quality under those conditions. But, for me, and given the present state of DSLRs, the flexibility of more pixels is worth the small tradeoffs involved.

 

[jrista]

Large pixels generally have high read-noise which means they tend to have a tough time in such environments. Read the text. They claim to have developed technologies that counteract that affect (that read noise increases with pixel size).

Hmm, ok. But in the general discussion, noise increases when pixels get smaller (at Hi-ISO).

Yeah, but that's a myth.

It is only a "myth" assuming images are always compared on a size-normal basis. That is certainly a valid way to compare, and the only normalized way to compare. However...assuming one buys a higher resolution camera for the purposes of using it for its native resolution, rather than downscaled to something smaller...the increased noise of a higher density sensor is no myth.

Smaller pixels have a lower cap on charge. Lower charge means higher gain. Higher gain means that for any given illumination level photon shot noise is exacerbated by amplification, which results in higher noise at native size.

Normalized is the way to compare that makes sense for... well... pretty much anyone. Saying that you should compare images at native resolution because one buys the camera to shoot at native resolution does not make much sense.

Sure, a shot from a hypothetical 40MP camera would produce a worse 36x24" print at higher ISOs (if you looked at it up close) than a 18x12" print from a 10MP camera with the same sensor tech. However, the 40MP print and the 10MP print would look almost indistinguishable if you printed them both at 18x12". However, and this is the big one, if you printed them both at 36x24", the 40MP print would look better at low ISOs (and basically no worse under other conditions).

(Sidenote: I was going to use 4x6" and 8x12" as example print sizes, but both sensors have plenty of resolution for that print size. 24x36 should be just about the smallest size where 10MP isn't quite enough)

You don't buy a camera just to shoot at native resolution (at least, I don't); you buy a camera to take pictures (or make prints). For example, I was doing relatively low light indoor sports shots last weekend with a 5d iii (typically f/2-2.8, iso 6.4-12.8k, 1/250-1/500). I had to downsize the images to about 3-5MP before I was happy with the quality (in other words, at higher resolutions I was either seeing noise or blur from noise reduction). So I could have gotten the same printable results with a 5MP camera in that particular case. However, the 20-whatever MP of the 5diii let me make that decision after I took the picture. And, if the light had been better, or if I cared enough to really buckle down with the post processing and noise reduction, the extra MP would have given me the ability to get higher resolution images.

Thats not to say that there aren't advantages to lower resolution cameras - a 5MP 5diii would not fill up the buffer nearly as fast, and previews would load faster in lightroom (the storage space isn't really an issue for me). A 10MP camera makes decent enough prints at 24x36 (certainly to the point where my skill, rather than the equipment, is the biggest limiting factor). And I am sure you could design a different sensor for every resolution and get slightly better print quality under those conditions. But, for me, and given the present state of DSLRs, the flexibility of more pixels is worth the small tradeoffs involved.

Well, for certain kinds of photography, everything you've said is probably true. I gather you do sports, which has some detail...with the finest being the players hair...which is not the primary thing viewers of your photography will be looking at. You can get away with a lot less resolution than many other forms of photography. Landscapes, for one, can not only use as much resolution as you throw at it, but near-infinite amounts of DR as well.

However, if you pick up bird photography, you'll quickly learn that no amount of sensor resolution is EVER enough, even though it means having to deal with more noise (and trust me, noise from smaller pixels is no myth, I deal with it every day). It is not as much a matter of how large you intend to blow something up as it is about whether you can adequately capture the extremely fine detail that exists in the scene. It's all about resolving power with bird photography, not whether you can eek out a 30x40 off 10mp. I really want to get my hands on a 5D III, a 600 II, and a 2x TC III, but even if/when I do, I'll still be getting a 7D II as well. There is just plain and simply no substitute for real-world native resolution, regardless of how large or small you indend to scale to in the end. Even with the best of the best gear, say a 5D III w/ 600 II & 2x TC III...you STILL fall short of the 7D w/ 600 II and 1.4x TC, both in terms of raw spatial resolution as well as reach (1200mm f/8 vs. 1344mm f/5.6). The problem is...despite the significantly greater detail the 7D right now *can* capture, ISO 3200 and 6400 are generally unusable. At native size, the smaller pixels just don't cut it in comparison to the 5D III, or even the 5D II for that matter....so I'm generally stuck at ISO 1600.

I don't even think that the 7D II, or any future 7D body, could ever actually quite perform as well as the 5D II or III. They pack in ~65,000 electrons at saturation with as little as 33% Q.E.. The current 7D tops out at just over 20,000 electrons and 41% Q.E. You would need a 120% Q.E. sensor (an impossibility, unless you aim instead for 80% Q.E. with color-splitting in place of a CFA) to achieve a 60k FWC on an 18mp APS-C and have the same native noise performance as a 5D III...and still be able to resolve enough detail for high-quality, professional bird photography (which these days, from the best of the best bird photographers, usually means a 1D X, a 600mm f/4 L II, 2x TC III, and a couple decades of skill in getting extremely close to your quarry. )

 

[Lee Jay]

Large pixels generally have high read-noise which means they tend to have a tough time in such environments. Read the text. They claim to have developed technologies that counteract that affect (that read noise increases with pixel size).

Hmm, ok. But in the general discussion, noise increases when pixels get smaller (at Hi-ISO).

Yeah, but that's a myth.

It is only a "myth" assuming images are always compared on a size-normal basis. That is certainly a valid way to compare, and the only normalized way to compare. However...assuming one buys a higher resolution camera for the purposes of using it for its native resolution, rather than downscaled to something smaller...the increased noise of a higher density sensor is no myth.

Smaller pixels have a lower cap on charge. Lower charge means higher gain. Higher gain means that for any given illumination level photon shot noise is exacerbated by amplification, which results in higher noise at native size.

Amplification has nothing to do with it and well capacity only matters at base ISO. And it's still a myth anyway as cutting pixel size by a factor of 16 (area) still allows double the resolving power at lower per-pixel noise in the final image because noise reduction software is so much better than spatial block averaging which is all bigger pixels do.

 

[jrista]

Large pixels generally have high read-noise which means they tend to have a tough time in such environments. Read the text. They claim to have developed technologies that counteract that affect (that read noise increases with pixel size).

Hmm, ok. But in the general discussion, noise increases when pixels get smaller (at Hi-ISO).

Yeah, but that's a myth.

It is only a "myth" assuming images are always compared on a size-normal basis. That is certainly a valid way to compare, and the only normalized way to compare. However...assuming one buys a higher resolution camera for the purposes of using it for its native resolution, rather than downscaled to something smaller...the increased noise of a higher density sensor is no myth.

Smaller pixels have a lower cap on charge. Lower charge means higher gain. Higher gain means that for any given illumination level photon shot noise is exacerbated by amplification, which results in higher noise at native size.

Amplification has nothing to do with it and well capacity only matters at base ISO. And it's still a myth anyway as cutting pixel size by a factor of 16 (area) still allows double the resolving power at lower per-pixel noise in the final image because noise reduction software is so much better than spatial block averaging which is all bigger pixels do.

Well, now your just cheating, throwing NR into the mix. Assuming a level playing field, and the use of NR for all sensors being compared, there is still more per-pixel noise for sensors with smaller pixels even after NR. With the kind of NR you would need to make ISO 3200 look acceptable, you would also lose a considerable amount of good detail, diminishing or eliminating the benefit of having a higher resolution sensor.

 

[Lee Jay]

You missed my point entirely. I'll spell it out.

You can cut the pixel linear dimension by a factor of four (giving you four time the resolving power), use NR to reduce the per-pixel noise to parity and still be left with twice as much resolving power.

 

[jrista]

You missed my point entirely. I'll spell it out.

You can cut the pixel linear dimension by a factor of four (giving you four time the resolving power), use NR to reduce the per-pixel noise to parity and still be left with twice as much resolving power.

No, I understood your point (although, I think you mean cut the pixel pitch by two, for four times the resolving power...you can fit four 1/2-pitch pixels into a 1-pitch pixel when factoring in two dimensions...if you divide the original pitch by four, you end up with 16x the resolving power, as you could fit 16 1/4-pitch pixels in the area of a single 1-pitch pixel.) I don't deny that you gain resolution...you do. But there is most definitely an increase in per-pixel noise, which limits how far you can push ISO on the high end before noise becomes such a problem that even with maximal noise reduction, you can't "have your cake and eat it to." You either have very noisy detail, or smooth "pixels" that barely or don't even offer the same detail as a lower resolution sensor. I wouldn't say it is impossible to get small pixels performing on a similar level as larger pixels...assuming you can get enough light incident on the photodiode, you could probably normalize the difference and make the higher resolution of an APS-C perform just as well as the lower resolution of a FF sensor...but we don't yet have that technology in a camera you can actually buy yet, nor will in the near-term future. Maybe if newer technologies quickly find their way into DSLR products, we might see better native-size performance out of an APS-C sensor...but as it stands now, even with NR, bigger pixels are still less noisy...by a factor of 2x or more. Assuming you normalize reach, fill the frame with your subject in both cameras, the higher image resolution of the 5D III means it would outperform the 7D on all levels, including detail and noise.