I think you're getting lost in the semantics and Brainiac isn't saying "density" in the way your reading it.
By all means, it sounds like Brainiac is describing "density" as aerial photosite density, or the percentage of whatever sized chip is actually used to gather photons (accounting for the microlenses, of course). For the following, I'm going to assume that both of Canon's most recent 1D's have the same fab processes beyond lithography to equalize any quantum efficiency. So for any cm^2 of sensor area, the same amount of light is getting turned into good data.
At equal chip size, a 1DM3-sized pixel should theoretically be a bit better on noise/area, due to the smaller amount of surface area taken up by the transistors and data lines, but I imagine that the microlenses reduce that difference significantly. Using the same argument the D3 should be a little better than either of the two 1D's on a per area basis, since it has even less of the chip's surface taken up by transistors instead of photodiodes. I would bet the effective difference is vanishingly small.
That said, the 1DM3 isn't the same sized chip as the 1DsM3, and cannot, under any circumstance gather more light than the 1DsM3--pure physics. D3 should be the "best" by a small margin, since it has the same total sensor surface area to the 1DsM3, and *slightly* less amount lost to transistors and data lines.
What I wrote is theoretically correct, but I could be putting the wrong words in Richards mouth.
That's absolutely correct, as far as I can see, Daniel.
In fact I am curious to see if the 1Ds3's larger area gives it the lead over the 1D3 at iso 6400 and 12800.
One other point: we know that Canon has been pushing high iso harder and longer than Nikon, and that Nikon has been doing heavier in camera NR than Canon, judging by the 5D, D200 and the rest. It is my opinion, and it's borne out by the correctly magnified DPR tree crops, that Canon actually still has a lead in sensors because the 1Ds3 without much built-in NR seems to match, at least, the D3 at iso 3200, and possibly higher, when the D3 clearly has quite heavy chroma NR built-in. The iso 12800 files I have seen from the D3 have looked quite plasticky and undetailed, so I think the actual sensor performance of the D3 is probably no better than the 1Ds3, and possibly worse. The D3 hysteria seems to me to be about good in-camera noise reduction, and the mere availability of those iso settings. So for me, because I am prepared to shoot raw, the 1Ds3 is the 'goto' for high iso.
What is more, this may have implications for Canon's next generation. They must surely include much more comprehensive noise-reduction in the Digic 4, in order to regain the benefit of their better sensors, and that will mean including much higher iso settings in all their cameras. The combination of something like Nikon's in-camera noise reduction, and a Canon full-frame sensor should make iso 25600 worth using.
As a little reminder of why I am not fully convinced by Chris's 1Ds3 v 1D3 comparison, here are the DPR iso 3200 crop from those two cameras again: http://cyberphotographer.com/1ds3/1Ds3v1D3.jpg
Judging by these crops, which may be unfair, after correct magnification, the 1Ds3 seems to better both the D3 and 1D3 at 3200 iso.
ChrisDM wrote:
You're talking semantics here obviously, but it doesn't change the simple fact that pixel density is a simple formula determined by number of pixels divided by the sensor area (size). Call it whatever you want.
You really do not want to stop trolling?
He has explicitly explained this to you _quite_ carefully twice now.
ulrikft wrote:
You really do not want to stop trolling?
He has explicitly explained this to you _quite_ carefully twice now.
Richard and I are having a civil debate discussing the semantics of physics, mathematics and electronics, and this is all you have to contribute to the debate? And who's the troll...? Feel free to add a productive point to the discussion if you are capable.
Here, let me teach you how its done. We were debating the definition of a "dependent" vs "independent" variable. To participate productively in the debate, I can refer to a link from Wikipedia on the subject:
Or I could just call you a troll. Which is the more productive, more mature path to take? I think I'll stick to the healthy debate and leave the name calling up to the children.
I think you're getting lost in the semantics and Brainiac isn't saying "density" in the way your reading it.
By all means, it sounds like Brainiac is describing "density" as aerial photosite density, or the percentage of whatever sized chip is actually used to gather photons (accounting for the microlenses, of course). For the following, I'm going to assume that both of Canon's most recent 1D's have the same fab processes beyond lithography to equalize any quantum efficiency. So for any cm^2 of sensor area, the same amount of light is getting turned into good data.
At equal chip size, a 1DM3-sized pixel should theoretically be a bit better on noise/area, due to the smaller amount of surface area taken up by the transistors and data lines, but I imagine that the microlenses reduce that difference significantly. Using the same argument the D3 should be a little better than either of the two 1D's on a per area basis, since it has even less of the chip's surface taken up by transistors instead of photodiodes. I would bet the effective difference is vanishingly small.
That said, the 1DM3 isn't the same sized chip as the 1DsM3, and cannot, under any circumstance gather more light than the 1DsM3--pure physics. D3 should be the "best" by a small margin, since it has the same total sensor surface area to the 1DsM3, and *slightly* less amount lost to transistors and data lines.
What I wrote is theoretically correct, but I could be putting the wrong words in Richards mouth.
Actually, I was the one who brought the term "pixel density" into the discussion, which is clearly defined as the number of pixels divided by the area of the sensor. There's just nothing more to it than that. All of these factors, plus many, many more play a role in how effective a sensor is at higher ISOs. That's the only point I've ever made, and will continue to make here. It baffles me that nobody else here has been able to comprehend this simple statement.
Daniel Heineck wrote:
At equal chip size, a 1DM3-sized pixel should theoretically be a bit better on noise/area, due to the smaller amount of surface area taken up by the transistors and data lines, but I imagine that the microlenses reduce that difference significantly. Using the same argument the D3 should be a little better than either of the two 1D's on a per area basis, since it has even less of the chip's surface taken up by transistors instead of photodiodes. I would bet the effective difference is vanishingly small.
That said, the 1DM3 isn't the same sized chip as the 1DsM3, and cannot, under any circumstance gather more light than the 1DsM3--pure physics. D3 should be the "best" by a small margin, since it has the same total sensor surface area to the 1DsM3, and *slightly* less amount lost to transistors and data lines.
What I wrote is theoretically correct, but I could be putting the wrong words in Richards mouth.
Yes, and your theoretical effective differences are exactly what many independent tests have shown. The most thorough being the review of the 1Ds3 just posted on dpreview today, which tests noise characteristics of all of the cameras and charts their performances against each other. It shows just what we've been talking about here, that throughout the ISO range both luminance and chroma noise being slightly lower in the D3 than the Canons, and the gap widening on the D3's performance at 3200:
Chris, I don't think anybody fails to comprehend your simple statement of the relationship between sensor size, pixel count, and density. I suspect most people reading this thread comprehended it long before you shared it. However many times you say it, I don't think anyone is going to have a Eureka! moment. I think that you have been called a troll, perhaps unfairly, for continually disputing the fact that pixel density is independent of sensor size by virtue of the third variable.
You also stated that pixel density was the 'most reliable first indicator' of high iso performance, a myth which this thread was intended to dispel, and a myth which your own generously contributed test helpfully does a great deal to undermine.
ChrisDM wrote:
Richard and I are having a civil debate discussing the semantics of physics, mathematics and electronics, and this is all you have to contribute to the debate? And who's the troll...? Feel free to add a productive point to the discussion if you are capable.
Here, let me teach you how its done. We were debating the definition of a "dependent" vs "independent" variable. To participate productively in the debate, I can refer to a link from Wikipedia on the subject:
Or I could just call you a troll. Which is the more productive, more mature path to take? I think I'll stick to the healthy debate and leave the name calling up to the children.
The thing is that Richard has explained very explicitly what he means by what he says several times. When you keep attacking that point, you come over as not very itnerested in having a productive debate. A 56x56mm sensor can have 10 pixels or 10.000.000 pixels, sensor size is not a direct indicator for the number of pixels or pixel size. Mathematically the pixel size IS a result of size/number, but that is not the issue here. It reminds me of a world of warcraft debate actually. People keep stating that you need a slow weapon to hit hard, when actually you need a hard hitting weapon to hit hard, slow is just a statistical bi-effect of the "hard" part, kind of like your dependent variable. It is mathematically/statistically kind of right, but the point is moot and not relevant in reality. That is why I wanted you to stop hacking it to pieces, as it looks quite ignorant from the outside. It looks like you _know_ that Richard has proven and shown that his initial theory was right, but that you WILL and CAN not realize this publicly and choose to cling on to an irrelevant and not very important fact to save some kind of internet face, I don't see this as mature, civil or rational in any discourse. I study law and I recognize logically sound and valid arguments when I see them, yours is not one.
As long as people keep linking to scientifically bogus tests, I am going to have to keep banging the drum.
As before, the comparison of the 1Ds3 and D3 here SHOWS CROPS AT DIFFERENT MAGNIFICATIONS, THEREBY SKEWING THE RESULT IN FAVOUR OF THE LOWER MEGAPIXEL CAMERA:
To make these results worthy of comparison, as if they are prints, it is necessary first to show them at the same magnification, and secondly to allow chroma noise reduction to be applied to the 1Ds3 file since it has clearly been applied by the D3 camera itself. I have also applied a gentle dust&scratches to the shadows which seems to give rough equivalence to the NR process the D3 applies. The order is 1Ds3 no NR, D3 at equivalent magnification, 1Ds3 with NR http://cyberphotographer.com/1ds3/dpr1ds3noisecrop.jpg
I also slightly adjusted colour balance and contrast of the 1Ds3 NR crops to match the Nikon's more closely.
None of this discussion has any merit when the debate involves comparing jpg images pulled from a web site. I find this especially ironic because Brainiac states he uses raw files exclusively, yet all the visual evidence is derived from in-camera processed jpg files, with each camera using a different lens.
A more accurate test would be to compare RAW files from the equipment being tested AND using identical lens on each camera body, with each raw file converted to JPG using it's own native software at default settings. Until then, this is a waste of time and bandwidth.
I don't use raw files exclusively. I am forced to use raw at 12800 iso because I don't have a Nikon.
I fully agree with you that a more accurate test would be great, but right now I only have access to one of these cameras. I appealed for a proper comparison to be done in the message which started this thread.
Meanwhile people continue to cite scientifically flawed tests (DPR) which don't set crops to the same pictorial magnification. What can I do but illustrate what those tests really show, as far as they go, since DPR staff are incapable of interpreting them correctly? One thing is for sure: my modified DPR crops here give a better idea than DPR's incorrectly magnified crops. I am only trying to show how wrong the DPR test method is, in the hope that people will do their own tests, recognising the fallacy of comparing crops at different magnifications, and demand a vaguely realistic test from their favourite test site.
I urge anyone who has understood that crops at different magnification (i.e. 100% crops from cameras with different numbers of pixels) are thoroughly misleading, to email any test site she frequents and request politely that they resize images to the same number of pixels before excising crops and publishing them to a deluded world.
AJ Nadershahi wrote:
None of this discussion has any merit when the debate involves comparing jpg images pulled from a web site. I find this especially ironic because Brainiac states he uses raw files exclusively, yet all the visual evidence is derived from in-camera processed jpg files, with each camera using a different lens.
A more accurate test would be to compare RAW files from the equipment being tested AND using identical lens on each camera body, with each raw file converted to JPG using it's own native software at default settings. Until then, this is a waste of time and bandwidth.
That was my methodolgy exactly when shooting this comparison between my 1Ds3 and 1D3. Same lens, same settings, same software, same conversion. The only thing that was slightly different between the two was focal length, which was necessary to compensate for the crop factor of the 1D3:
I'm not one to normally make such academic comparisons, but I thank brainiac for sparking me to make this one. I was pleasantly surprised to see how well my 1Ds3 competed at 3200 versus my 1D3. It didn't change anything for me though, the 1D3 is still the obvious choice for low light/event shooting, but it is nice to know the big boy can hang...
I get a very similar result with my 5D at ISO 3200 "pushed," although I'm not 100% certain it would resolve those 3 lines at the bottom of the guitar. I'll have to look over some of my ISO 3200 files a little more closely for very fine detail.
I recently photographed the Anat Forte jazz trio at the Blue Note in New York City. All were ISO 3200 with a Canon 28-70L and 70-200L IS. I'll take a look and maybe post.
ulrikft wrote:
The thing is that Richard has explained very explicitly what he means by what he says several times. When you keep attacking that point, you come over as not very itnerested in having a productive debate. A 56x56mm sensor can have 10 pixels or 10.000.000 pixels, sensor size is not a direct indicator for the number of pixels or pixel size. Mathematically the pixel size IS a result of size/number, but that is not the issue here. It reminds me of a world of warcraft debate actually. People keep stating that you need a slow weapon to hit hard, when actually you need a hard hitting weapon to hit hard, slow is just a statistical bi-effect of the "hard" part, kind of like your dependent variable. It is mathematically/statistically kind of right, but the point is moot and not relevant in reality. That is why I wanted you to stop hacking it to pieces, as it looks quite ignorant from the outside. It looks like you _know_ that Richard has proven and shown that his initial theory was right, but that you WILL and CAN not realize this publicly and choose to cling on to an irrelevant and not very important fact to save some kind of internet face, I don't see this as mature, civil or rational in any discourse. I study law and I recognize logically sound and valid arguments when I see them, yours is not one. ...Show more →
Wow, now those were some useful comments. Doesn't it feel better to participate rather than call names? I thought so.
Regarding your World of Warcraft debate, kinetic energy (or "hitting hard" as you put it), in terms of weapon trajectory is a function of speed times weight of the trajectory. So would that make kinetic energy or weight the dependent variable? Does it matter what we call it? Just like the Pixel density equation, you can move the 3 variables around the equation any way you want. You can fix any of the 3 variables in the equation to a set number (the independent variable), and the remaining two variables will be the dependent variables (move one up, the other goes down, etc). Richard's argument is based on the sensor area being the independent variable. And as number of pixels on that sensor increase or decrease, the pixel density also moves inversely proportional to it. You could also arrange the equation to make the sensor area the dependent variable, and fix the resolution (the independent variable). So as sensor area decreases, pixel density increases. Once again, just semantics. But what does this have to do with the topic, which is "high ISO"? That just depends on whether or not you believe pixel density affects noise. And I guess the jury's still out on that one...
...er, pixel density doesn't become a dependent variable (dependent on sensor size) until you specify that the number of pixels is a constant. We are not specifying that. I think you are mis-applying the fairly loose wikipedia discussion of this.
brainiac wrote:
...er, pixel density doesn't become a dependent variable (dependent on sensor size) until you specify that the number of pixels is a constant.
Right. There are 3 variables in the equation, one independent variable and two dependent variables. You can make any one of them the constant, independent variable, then the other two are inversely dependent on each other (as one goes up, the other goes down). But as I pointed out two pages ago, it's just semantics, not sure how we got so hung up on it.
ChrisDM wrote:
Right. There are 3 variables in the equation, one independent variable and two dependent variables.
I don't think so. There is only a dependent variable IF you set one of the variables as a control variable. Density is only dependent on size if pixel count is constant, let's say 12 megapixels. In considering the 1Ds3, D3, 1D3, and G9, none of the variables is constant. Our real dependent variable is noise per square inch of print, i.e. the 'image to noise ratio'. It is primarily dependent on sensor size, and largely independent of pixel density. That's why high iso noise from 1Ds3, 1D3, and D3 looks so alike, and quite unlike the ouput from compact cameras, which look mostly the same, whether 6 megapixel or 12.
brainiac wrote:
I don't think so. There is only a dependent variable IF you set one of the variables as a control variable. Density is only dependent on size if pixel count is constant, let's say 12 megapixels. In considering the 1Ds3, D3, 1D3, and G9, none of the variables is constant. Our real dependent variable is noise per square inch of print, i.e. the 'image to noise ratio'. It is primarily dependent on sensor size, and largely independent of pixel density. That's why high iso noise from 1Ds3, 1D3, and D3 looks so alike, and quite unlike the ouput from compact cameras, which look mostly the same, whether 6 megapixel or 12....Show more →
The equation I'm referring to, which I clearly stated many times, is pixel density.
Gosh guys I tried this out (as previously posted) and the Mk III and 1Ds Mk III have really identical Hi ISO noise for the same image-just like brainiac sez.
And no you don't use the identical lens because then you don't have the same image. The 35 on the 1Ds Mk III and the 28 on the Mk III result in the same image sure the 1.274X Mk III's crop sensor "Crops" but that doesn't mean that you must enlarge either image file. Only that the crop sensor does mean that you use the lens FL necessary to frame the image you want.
The noise pattern @ 500% is almost identical. The difference is due to 1Ds Mk III's double pixels. But that's @ 500%! Not where you could ever use/see it.
So you can waltz as many angles on the head of a pin as you desire but the actual image result is really where the rubber meets the road.
