Canon better get the ball back - I'd hate to have to wait for the 5D MIII before I upgrade again!

http://www.kenrockwell.com/nikon/d3/iso-1.htm

It's a bit puzzling that Nikon's cropped-format latest offering, the D300 is so much less sharp than the D3, yet they both have the same number of pixels. The explanation is detail-destroying noise reduction, but as far as I know these differences do not exist between the 5D and the 40D and I'm led to believe that the D300 at ISO 3200 is at least as sharp as the 40D and considerably less noisy, according to Image Resources comparator.

So something is seriously amiss here. Below are 200% crops of the D300, 40D and E-3 taken from Image Resources comparator.

Click to view attachment

Love the line from the article "instead of using genuinely the best lenses, I used the lenses that people who don't actually own them think are the best"

Also, wondering what advantage the 105mm macro provides over the 70-200, if any.

"I see the D3x as the clear winner. It's clean and sharp. The fur looks like fur and has real texture everywhere, as well as being clean. No noise and real texture means high signal to noise ratio, and the winner!"

Ken Rockwell is a goof. While I found the crop comparisons pretty insightful, his conclusion that the D3 "trashes" the 5D is a little hasty. I would say that the D3 has a marginal edge straight out of the camera, because it seems to use more chroma noise reduction than the 5D. But chroma noise is very easy to deal with in RAW, and considering that the D3 seems to preserve the same amount of fine detail, I would say that the cameras are about equal. There's obviously more luminance noise reduction going on in the D3's default setting as well, but I think it's handled FAR more judiciously than with the D300. I would also like to point out the different points of focus in the D3 crop and the 5D crop directly below it: the 5D is focused more towards the very front of the bear, towards his eye, and the D3 is obviously more focused to the back, where the beige fur meets the brown. I believe that their different points of focus hinder, but don't prevent, a comparison of fine detail. Considering this, I don't see any profound difference between the two cameras in their ability to capture fine detail.

I do, however, think Nikon's efforts to catch up with Canon (Full frame only) is an impressive feat, but these crops don't provide enough evidence to say that Nikon's D3 trashes the 5D. I think it's equally impressive that the older 5D is still able to compete with a newer camera with respect to fine detail at high ISO. I believe that Nikon's next generation of CMOS cameras will be more impressive than this first generation.

FYI, I don't own either camera.

despite the 5d being what, 2.5 years old, and still holds up pretty good to the newest kid on the block.

Nor I, but after examining RAW data from both cameras, the D3 is clearly better than the 5D in every noise issue; the only thing inferior about the D3 (and this only applies to some types of photography), the D3 clips the RAW data at black, which means that the extreme, deepest shadows are somewhat compromised. Some have claimed that the clipping is actually above true black; I haven't checked that myself.

The D3 collects about twice as many photons as the 5D with the same exposure, and has almost a stop advantage in read noise at ISO 1600, and 1/2 stop at ISO 100. Unless you're doing astrophotography stuff or its equivalent, the D3 should do everything the 5D does (noise-wise), and better.

Now the D3 vs the 1DSmk3, that's another story. The D3 may collect more photons, but the read noises at the pixel level are similar for both cameras, and the 1DSmk3 has almost twice as many pixels, so the image-level read noise of the 1DSmk3 is clearly the best in the industry.

Here's a 100% crop of the 1DSmk3 at ISO 1600, from a RAW file linked to at LL in another thread. This is my own manual RAW conversion done in IRIS, no processing other than white balance and color interpolation (no NR whatsoever; less than "zero" NR in converters), taking a deep shadow area and pushing it to ISO 51,000 (this image's DR spans only 250 14-bit levels):



Squint and look at it again, to simulate software NR. Then imagine this extended to the entire image (3 to 4 monitors wide and 3 to 5 tall).

deleted

Excellent.

The UK magazine Amateur Photographer has a review of the D300, and 100% crops at ISO 1600 and ISO 3200 are amazing, and show very very little detail loss.

There are also online comparative D300 reviews with samples images which agree with AP but contradict Ken's results.

I think what Ken has done is choose some D300 settings with too much in-camera noise reduction.

In other words, it is a bit of a pointless test. Now if he had played around with the settings for each camera, to optimise the JPG IQ, that would have made sense. But it looks like he just rushed in head first without thinking too much.

Yeah! That's the impression I get. I think the D3 might be marginally better than the 5D regarding high ISO noise and resolution, but that D300 seems to trounce the 40D. It's more expensive of course and in a sense you get what you pay for. However, if Canon had given us a 40D with the specs of the D300, I'd be buying it without doubt.

1.) How do you mean by 'image-level' read noise?

1a.) Why is it that higher pixel count will decrease image-level read noise?

The intensity of noise when equalizing all images to being viewed at the same size, from the same distance.



Because then each pixel is a smaller size in the observer's FOV, and the intensity of noise is dependent upon the displayed pixel size.

Try this; create a rectangle of 50% grey in photoshop. Now, add 2% chromatic gaussian noise, and take a look. Now, go to the Navigator tool, and increase the zoom from 100% to 200%, 300%, etc. The noise level, from the pixel-centric viewpoint, does not increase as you are doing it, but the bigger you make each pixel, the more intense the noise becomes. Try it again, standing 10 feet from the monitor, by having 100% in one window, and duplicates at 200%, 300%, and 400% in other windows (or get someone else to zoom for you with a single window, or use a remote mouse and keyboard).

IQ-wise, the 300D only "trounces" the 40D in terms of 20% more pixels. The 300D has more noise than the 40D, both shot and read, at all ISOs. It's all software, Ray; all software. Hiding noise has been a priority of companies that have historically had relatively noisy cameras. Canon does little to hide theirs, because they haven't had as much. Canons could just as well sport ISOs 12,800, 25,600, etc, with aggressive NR. It is not Canon's style to use aggressive NR, and the public is naive about noise, so they don't offer them.

The big problem, right now, in terms of people judging cameras' capture performance is that most people don't recognize noise reduction, and don't seem to notice the missing real detail, and are happy with that substituted painterly look. To me, IQ is about being able to see what was really there, not about "not seeing noise (or grain)".

Draw cartoons. They have excellent low-noise characteristics.

Duplicate post erased

This duplicates my own experience completely, as I mentioned in the other forum following from Michael's tests.

I really don't think that pixel level comparisons are as valid as some think they are.

We don't wiew pixels, we view prints (or images on a monitor, etc).

What is the point of viewing the noise of a 20 x 30 print from a 1Ds II vs the noise of a smaller print from a D3? There is none.

I've read a couple of reviews that say that the D3 image can be magnified to the size of that from a 1Ds, and it will look sharper, but their own examples don't bear that out.

Since the vast majority of photography will be taken at 1600 and below, usually at 400 and below, the difference at higher ISO's is minimal.