EOS 5D and 24-105 f4L IS USM
EOS 20D and 17-85 f4-5.6 IS USM
PART 3 - noise at high ASA (24 September 2005)
The sensor inside the EOS 5D is twice the size of that in the 20D. Twice as much light falls on it. However the resolution is not double that of the 20D, only 50% more, because each pixel on the 5D is bigger than each 20D pixel. Assuming all else is equal, this ought in theory to give the 5D a better signal to noise ratio than the 20D, as each pixel is gathering more light. In other words, the EOS 5D images should look less noisy.
I have tried to set up a realistic low-light test. In practice I would only use high ASA when it is necessary to be able to hand-hold the camera, ie when it is dark. Using high ASA in bright sunlight may well give very low noise, ie good signal to noise ratio, because there is plenty of signal there. But in real life we use high ASA only when we need to, such as when shooting wide open at 1/25th of a second as with this example.
The difference, even at 100% magnification (shown above) is quite small at 1600ASA. Both cameras have produced a perfectly useable result, and in both cases the noise could be reduced using software.
At 3200ASA, it's a different story. Both results are softer and noiser, but the 5D is clearly the better of the two. Could we say the 5D is comparable at 3200ASA with the 20D at 1600ASA? Here's the side-by-side comparison:
Torture test - 6400ASA!
How, you may ask, do I propose to test these cameras at ISO 6400ASA when the maximum setting (and even then only when enabled through a custom function) is 3200ASA? Easy - set the ISO to 3200ASA and dial in one stop of underexposure, giving an effective 6400ASA, then add back one stop of exposure when converting the RAW file.
Things are getting pretty manky for both cameras at 6400ASA, although at least in the case of the 5D, perhaps not so bad considering this is a 100% crop of a small area, and has not had the benefit of noise reduction software such as Neat Image or Noise Ninja. Even without such software, there is a simple technique published recently on the Luminous Landscape here which can work wonders. I imagine noise-reduction software and techniques will continue to improve, so in a few years time we can always go back to noisy RAW files and produce better results. I think it would have been more sensible for Canon to include 3200ASA as a proper setting and having the "H" setting as 6400ASA.