Larger photosites means less noise?

According to this well done test, it doesn't!

http://www.sphoto.com …rsensors/dslrse​nsors4.htm

Very strange indeed!

Another strange thing I found, check the Luminance Noise Graph:

http://www.dpreview.co​m …canoneos1dmkii/​page18.asp

1Ds is noisiest of them all? Even compared to the CCD 1D? And also D2H?

Can it be explained at all?

It's not entirely clear from his write-up whether he compared pixels at 100% crop. Certainly, he only showed same-area crops, which may, or may not, depending on how much running back and forth he did with his camera, correspond to equal numbers of pixels. In particular, the greater pixel densities of the 20D and 1Ds II compared to the 1D II for the same areal crop will tend to mask noise. If he'd taken 100% crops of each of the three and compared the noise, the results might have looked different. If you've got (pulling numbers out of thin air) 100 pixels in one frame, 169 in the second and 256 in the third, unless you can resolve the individual pixels in all three, your eye will average out any noise in the shots with sub-resolution pixels.

He also did a wall shot at 15" distance (on all 3 I assume since FOV doesn't factor in) and reached the same conclusion. Weird.

Again - the pixel density for equal area crops would be different. The 1D 2 has bigger pixels than either of the other 2, so fewer pixels per unit area, so easier to see noisy pixels.

Oh... so to compare noise levels, one must crop equal amounts of pixels both horizontally and vertically? If so, got it!

Yep! If you want to pixel peep, you really need to peep at the individual pixels! No fair comparing the average of 25 with the average of 4.

All other things being equal, a larger sample size will get you more accurate results. For instance, a rain gauge that is 12" in diameter will give you a more accurate measure of rainfall than you could achieve if you used a straw. (However, the straw still might render fairly accurate results in certain instances.) Photosites would work the same way. The larger photosite should more accurately measure the incoming light compared to a smaller site.

However, changes in technology (IOW, all other things not being equal) can change the equation. Newer but smaller photosites could be more sensitive and more accurate than their predecessors, even the larger ones. Advances in materials could allow the structure that holds the photosites in place to be reduced so that the photosites could be placed closer together (with less non-photosensitive structure in between them) resulting in more light being measured within a given area. Etc.

IOW, I think it is an oversimplification to only look at the size of a photosite.

In this particular case, the person running the test was comparing three cameras of approximately the same vintage - the 20D, 1D II and 1Ds II. The primary variable was the size of the photosite.

Wait till someone does a proper test. dpreview seems to be the most sensible at these things

Howdy, stranger! Where ya been?

Really flawed test. From that site:

I underexposed the image about 3.5 stops so the resulting image would only show 'dark noise'.

Talk about LOOKING for noise! And then that guy pushed the image 3.5 stops with ACR. Please try a real test with correct exposures.