Like the EOS 30D, the new Canon EOS 40D allows ISO to be set anywhere from 100 to 1600 in 1/3 or full ISO steps. The 1/3 sequence is 100, 120, 160, 200 etc., while the full steps are 100, 200, 400, 800 etc. There is also an "H" setting which can be enabled via a custom function which enables the camera to operate at an effective ISO 3200. You can't select 1/3 ISO steps between 1600 and 3200 though.
New on the EOS 40D is a high ISO noise reduction mode which can be selected via a custom function (CFn II-2). The downside of selecting noise reduction is that the maximum burst length when shooting in high speed continuous mode will be reduced (see EOS 40D manual, p156).
As you can see, the best match for the image shot with the 20D at 1/4000s is that shot with the 40D at 1/3200s (20D luminance 122; 40D luminance 123). The image shot at 1/4000s is a little darker (luminance 105 vs. 122) and the shot at 1/5000s was darker still (Luminance 91 vs. 122). This would indicate that the ISO settings on the EOS 40D are about 1/3 stop slower than the same settings on the 20D.
In a more practical test I shot three different scenes with both the 20D and 40D, using the same ISO setting (400), the same metering mode (Evaluative) and the same auto exposure mode (Aperture Priority). The results are shown below:
In the top image (red car) the 40D chose a shutter speed 2/3 stop slower than the 20D and
produced a brighter image. In the center image (tree and sky) the 40D chose a shutter speed 1/3 stop slower than the 20D and the images are quite similar in brightness. In the lower image (grass in sun and shade) the 40D again chose an exposure 2/3 stop slower than the 20D and the image is brighter. These differences in exposure and brightness appear to come from two causes. First the 20D is about 1/3 stop faster in terms of ISO, so for the same image brightness the 40D would give 1/3 stop more exposure. This is shown in the center set of images. However the metering system of the 40D seems to allow for an additional 1/3 stop of exposure in some cases (top and bottom images), resulting in a brighter image. In no case are the highlights "blown out", so the extra exposure of the 40D is useful in recording more shadow detail without any loss of highlight detail. Technically speaking the exposure of the 40D is probably slightly closer the the "correct" exposure than the 20D. In terms of the histogram function, the 40D (in evaluative metering mode) seems to expose slightly more "to the right". Note that these tests were all done at ISO 400. It's possible that there may be differences at other ISO settings and if I see them, I'll update this section of the review.
[• End of update 09/21/07]
On the basis of the few test shots I've taken up to now I'd say the high ISO noise (at "H" or ISO 3200) seems a little higher in the 40D than in the 20D images with the High ISO Noise Reduction function turned off. However with the High ISO Noise Reduction on, the 40D images show lower noise then the 20D images with no apparent loss of sharpness. This means that, with noise reduction on, at ISO 3200 the 40D delivers cleaner, sharper images than the 20D (and presumably 30D too). The high ISO noise reduction function seems particularly effective at reducing chrominance noise.
The images below show these effects. They are 400% crops from the originals, i.e. they have been enlarged 4x to show greater detail, in fact you can start to see the individual pixels. There really seems to be no observable loss of detail with noise reduction turned on.
These images were shots as the highest resolution and quality JPEGs. How they would compare shot RAW I don't yet know. I will be doing some more controlled testing later, looking at the whole ISO range, but for now I'd say the 40D high ISO performance is good, especially with the noise reduction function turned on.
Here's more of a "real world" shot taken at ISO 3200 (high ISO noise reduction on):
and here's the section outlined in yellow at 100%. On my monitor (17" with 1280x1024 screen resolution) this represents a crop from a print approximately 36" x 24" - a pretty big print!
I think it's very good performance for a JPEG straight out of the camera at ISO 3200.
I've had the chance to do some more analysis of the noise issue, looking at all speeds from ISO 100 to ISO 3200, using both color and greyscale charts and analyzing the images using Imatest and Noise Ninja. Basically my initial observations seem to be confirmed. The 20D shows slightly lower noise than the EOS 40D with high ISO noise reduction off, but slightly more noise then the EOS 40D with high ISO noise reduction on. High ISO noise reduction works at all ISO settings and mainly seems to drastically reduce chroma (color) noise. It doesn't have a big effect on luminance (intensity) noise. As with most cameras using a conventional Bayer color matrix (2G, R, B), noise is lowest in the green channel and highest in the red channel.
Besides the absolute amount of noise you have to also take into account that for a given size at a given ppi setting, the 20D image has to be enlarged about 11% more because it contains fewer pixels. This tends to make any noise more visible.
Below are a set of color test patches shot with the 20D and the 40D (noise reduction on and off). The 20D image has been scaled to match the 40D image to give a better visual impression of what the noise would look like in a given size print.
These images have not been altered (no color or density corrections) other than to change the scaling of the 20D patches from 100% to 112%). The exposure and colors are what the cameras recorded (Large/Fine JPEGs).
Again I'd conclude that the EOS 40D noise is very close to that of the EOS 20D, which is pretty good. Analyzed on the level of the actual noise level at the sensor (ignoring differences in sensor size), the 20D seems very slightly lower than that of the 40D (high ISO noise reduction off), but on a practical basis based on equal sized prints, any differences are small enough to be of no importance. Turning on high ISO noise reduction lowers the noise to below that of the 20D, mostly by reducing chroma noise.
Though I've only included examples at ISO 3200, where noise and noise differences may be
most visible, the same conclusion hold for lower ISO settings, though it becomes increasingly more difficult to visually detect noise or differences in noise as ISO is lowered.
[end of 09/08/07 update]
It's always a little tricky giving dynamic range numbers since it depends on your definition. Measuring dynamic range on the high side is fairly simple since the point at which the sensor saturates (values go to 255) is quite distinct. However on the low (dark) side you have to define what's actually useful. The pixel value will eventually fall to zero, but it does it slowly and mixed in with noise. So if you have a pixel value of 2, you haven't bottomed out (a pixel value of zero), but you might have +/- 1 units of noise and it's very unlikely that in any print you'd be able to tell the difference between a 0 and a 2 value and if you boost the levels via contrast or histogram shifts, you'll boost the noise along with the signal and get very noisy data.
All of that being said, here are some dynamic range measurements I made on JPEG files:
The plots are for ISO 100 and ISO 3200 (highlight tone priority off) and ISO 200 (highlight tone priority on). It's clear that there's really very little change in the curves going from ISO 100 to ISO 3200 with HTP off. On the plus side (highlights) you have around 3.5 stops. On the minus side there is data above zero down to about -6 stops, but as I discussed above, at -6 stops the data really isn't very useful, especially at ISO 3200 because of the noise. I'd say there's maybe 5.5 stops at lower ISO settings and maybe 4.5 stops at ISO 3200, depending on how much noise you can stand!
The plot for ISO 200 with HTP (highlight tone priority) on does show that it gives you a little more dynamic range on the plus side, especially between about +2 and +4.5 stops. This is consistent with the detailed HTP tests I describe in section 6 of this review.
If you shoot RAW, you may be able to extend the DR by up to a stop at each end of the range, but the quality of the data may be questionable, especially at the high end if one of the color channels is clipping.[end of 10/23/07 update]
Past Canon EOS DSLRs have always generated warm images under tungsten lighting, even with the white balance expressly set to the "tungsten" setting. The reason is said to be that people like the warm look it gives, and expect it from images since that's how film would often render scenes shot under room lighting.
Based on a few test shots, the white balance of the 40D seems pretty similar to that of the 20D, which generally good performance in most modes, but a warm bias under tungsten lighting.
As you can see, auto white balance gives a very warm result. Tungsten white balance is closer to neutral, but still warm. Custom white balance gives a neutral grey - as it should. In fact it reads 130:131:131 for the R, G and B channel intensities. A white balance of 2500K gives a cool (blue) tone and a 300K white balance gives a slightly warm tone, very similar to the tungsten setting.
[end of Update 09/10/07]
NEXT-> Part IV - Autofocus