Sony A900: A Closer Look at 24.6MP Resolution and Noise
by Wesley Fink on October 27, 2008 2:00 AM EST- Posted in
- Digital Camera
Full-Frame and 24.6MP Compared
The A900 is the first Sony to feature a full-frame DSLR. The sensor is approximately the size of a frame of 35mm film, which is 24x36mm, and 24.6MP is the highest resolution currently available in a full-frame camera.
Compared to most digital cameras today where the sensor size is closer to APS-C size, the lenses appearing to be 150% to 200% longer than the specified focal length. Other Sony DSLR cameras, like the A700, feature a 23.5x15.6mm sensor. 24.6MP is also double the sensor resolution of today's prosumer or advanced amateur models, which are generally around 12MP.
In the computer world smaller and smaller traces mean higher density, more transistors, and generally better and faster performance. However, the digital sensor is not a digital device. Instead, it is an analog device that gathers light and turns it into a digital signal. The reverse is true in sensors in that larger sensor size is almost always better, with everything else equal.
As you can see in the chart below, none of the APS-C sensors is even half of the area of a full-frame sensor. With a range of 28.1% to 42.4% of full-frame size, there is clearly a lot more information that can potentially be captured with a full-frame sensor.
| DSLR Sensor Comparison | ||||
| Camera | Effective Sensor Resolution | Sensor Dimensions and Area | % of Full-Frame | Sensor Density (MP/cm2) |
| Olympus E-520/E-3 | 10.0 | 13.5x18 2.43 cm2 |
28.10% | 4.0 |
| Canon XSi | 12.2 | 14.8x22.2 3.28 cm2 |
38.00% | 3.7 |
| Canon 50D | 15.1 | 14.9x22.3 3.32 cm2 |
38.40% | 4.5 |
| Pentax K20D | 14.6 | 15.6x23.4 3.65 cm2 |
42.20% | 4.0 |
| Sony A700, Nikon D300, Nikon D90 | 12.3 | 15x23,5 3.66 cm2 |
42.40% | 3.3 |
| Sony A350 | 14.2 | 15.8x23.6 3.72 cm2 |
42.90% | 3.8 |
| Nikon D700/Nikon D3 | 12.1 | 24x36 8.64 cm2 |
100% | 1.4 |
| Canon 5D | 12.7 | 24x36 8.64 cm2 |
100% | 1.5 |
| Canon 1Ds Mark III | 21.1 | 24x36 8.64 cm2 |
100% | 2.4 |
| Sony A900 | 24.6 | 24x35.9 8.61 cm2 |
100% | 2.9 |
The last column in the chart is the one that tells the story most accurately, however. Here the resolution of the sensor is divided by the sensor area to yield a sensor density. The lower the density, the larger the individual pixel size, and the more info that pixel can gather - all else being equal. There are a few surprises here, such as the Sony A350 being essentially the same density as the Canon XSi, and the new Canon 50D having the highest density of any current DSLR camera.
The last column does put into perspective the true potential of the full-frame sensor and sheds some light on the true meaning of Sony's 24.6MP A900 sensor. At 2.9MP per cm2 the A900 still exhibits a lower density and theoretically better high ISO performance than any current APS-C DSLR. This is very much at odds with the ridiculous claims many on the web are making about Sony going too high in resolution on the A900. In fact, sensor density on the A900 is lower than the 10MP Canon 40D, which is 3.1.
The point is that any issues the Sony may be found to have with noise are not the result of pixels being "too small". All else being equal the high ISO noise should be at least as good as an 8 to 10MP Canon sensor. Where the Sony does suffer is in comparison to sensor density of other full-frame sensors. In that metric the Sony has twice the pixels per cm2 of a Nikon D3/D700 and Canon 5D, and keeping up in high ISO performance with those cameras would be quite a feat.
Facebook
Twitter
RSS
45 Comments
View All Comments
CEO Ballmer - Monday, October 27, 2008 - link
These things are very nice, but the ZuneCam will prove they are waaaaay over-priced!http://fakesteveballmer.blogspot.com">http://fakesteveballmer.blogspot.com
strikeback03 - Monday, October 27, 2008 - link
I'm guessing default settings in JPEG. Which is a valid baseline, but given different manufacturers different philosophies on JPEG processing, makes commenting on apparent sharpness of the image somewhat difficult. Also, was any form of dynamic range expansion turned on? I'm surprised at how much color noise is already appearing in the A900 shots at ISO 400, wonder if it is a result of boosting some shadows?As you guys are buying your test bodies, I can understand not having a 1DIII, but making statements such as "At Lower ISOs, 100-400, nothing on the market comes close to the Sony." seems a bit odd without at least trying it against a body with 86% of the pixel count. And obviously if those in the market have the means to get a medium-format back and associated gear, you are far beyond the capabilities of the A900.
Wesley Fink - Monday, October 27, 2008 - link
Yes, default settings were used in all parameters, including dynamic range expansion. We tried to be careful to say compared to others in its class, which is obviously full-frame 35mm-size sensors. We have not compared any of the full-frame cameras to medium forma or anything like the new Leica medium format that uses the Kodak sensor, nor do we plan to.As for comparisons to the 1Ds Mark III, the comments were based on conversations we had with Pros who have shot both the Sony A900 and the Canon 1Ds III. We also looked at images shared with us by a couple of them. We would be more comfortable with our own hands-on with the $8000 1Ds III but that will be a moot point once we have the 5D Mark II. Canon claims the 5D II sensor, with the same resolution, is superior to the 1Ds III so we will soon get our first hand look.
melgross - Monday, October 27, 2008 - link
Some odd conclusions both here and DPR.One thing that stands out here though, is the strange notion that with all the noise, the higher resolution is of any use.
I don't know of anyone who wouldn't control noise with noise reduction once it reaches the levels of the a900 at ISO 800. Even 400 isn't great, just ok.
Once that is done, the higher resolution will be no better than anywhere else. The more noise, the more noise reduction, the less detail.
It would have been more interesting if noise reduction was applied to see the effects of bringing it in line with the other cameras. Where would the detail have gone?
In addition, if noise is so high at higher ISO's, as it is, making smaller prints so as to not see it, would also mean making smaller prints than would allow you to see the higher resolution as well. No gain, no pain.
I also find it interesting that field reports about this camera from both DPR and the Luminous Landscape have agreed that real-world pictures show less detail that the Canon 1Ds mkIII. The high detail is smeared because of the non removable noise reduction done before the A/D converter (as opposed to the noise reduction that can be turned off after the A/D converter.
The subject used for the comparisons is also pretty bad. I'm sure something more suitable that that could be found.
JarredWalton - Monday, October 27, 2008 - link
I shoot plenty of objects at ISO 100-400, sometimes using a tripod. In such cases, the differences between 12MP and 26MP will definitely be apparent. That said, I don't do print work and will just end up cropping and resizing down to a manageable resolution. Still, for details of a motherboard as an example, that increased resolution could be very handy.As for the subject matter of the ISO comparisons, I'm of the opinion that as long as it's static (i.e. the same in each review) it serves its purpose. We've all seen the various chart photographs (which are widely available elsewhere), but I suppose Wes can reproduce those if necessary. I just don't see how it makes much of a difference one way or the other; the sample images at the end do a good job of showing other possibilities with the camera, and personally I think some of the images look very nice indeed. Just stay away from ISO 800 or higher if possible.
strikeback03 - Monday, October 27, 2008 - link
My complaint with the ISO image subject is a lack of anything to help differentiate between good NR software and a low-noise sensor. With a test image composed almost entirely of smooth single-color areas, good NR software could make those pretty smooth while maintaining the borders between colors. A subject with more detail makes it harder to just blur noise away without seeing the results in detail captured.haplo602 - Monday, October 27, 2008 - link
and on a color checker chart you can evaluate sensor performance in high ISO for colo/saturation drifts ... this cannot be seen on the current simple 3 color test target ...LTG - Monday, October 27, 2008 - link
This review kept reminding me of the DPReview.com review of the same camera.It's definitely different works, but the emphasized points are very similar like choosing to commend sony for lack of gadgetry.
I guess that fine we all build thoughts from some starting point - but I'll eat my left shoe if the author didn't read the dpreview article before writing this.
yyrkoon - Monday, October 27, 2008 - link
You know, you could have just as easily not left a comment at all.No one really cares what your thoughts are on web plagiarism . . .
Wesley Fink - Monday, October 27, 2008 - link
Plese read my Sony A900 Preview where I praise the same points long before dpreview published their review of the Sony A900. We all read other reviews but I have no problem reaching my own conclusions.