Nikon D3X and Sony A900 ISO 1600 comparison

[David Kilpatrick]

So is that a 70-200 lens on the Nikon? It is not as sharp as the Sony
but the Sony looks more noisy, in particular color noise at the bottom
comparing the black neck & white strings.

No, both lenses are consumer level - 28-105mm on the Sony is a 1993
Minolta RS zoom, 24-120mm on the Nikon is the latest VR. I have also
been testing the Canon 5D MkII and they supply the 24-105mm L f4. I had
to request further lenses, as the 24-105mm had such serious defocusing
to one side of the image. I asked Nikon for the 24-120mm to see whether
it is substantially better than the Canon 24-105mm. So far, it appears
to be generally rather soft all over, but not suffering from decentering
or VR/IS problems. The Canon is much sharper centrally, but that's not
much use on full frame if the corners are very soft and one side of the
image tends to be smeared by some kind of decentering (probably caused
by the IS system).

I've also tested the Canon with a 50mm f1.8 and 100mm macro, and of
course, it's perfect with these. Nikon is due to send me the new 50mm
f1.4 next week.

If it helps for perspective, I sold my Minolta 24-105mm and the much
older 28-105mm is a replacement - I knew it was sharp enough before
doing the test. The newer 24-105mm simply was not a professionally
acceptable lens - sharp in the centre, but too much distortion and too
much CA for a high res full frame camera. This, sadly, is the lens which
Sony is selling with the A900, in a new skin. And Canon is doing 5DMkII
kits with their 24-105mm. Seems that the makers don't bother to test
their gear fully, or they believe the myth that all their products are
perfect!

David

[David Kilpatrick]

The focal length specs are measured when the lens is focused at
infinity. There's nothing different about the lenses and this isn't a
new effect either. That being said, all those lenses should have
relatively the same FoV when focused at infinity.

True - internal focus could result in the Minolta 28-105mm being 95mm at
a fairly close focus. But the relative values are inconsistent across
two macros and three zooms, from today's shooting. The balance of errors
is probably down to how the lenses are chipped for reporting focal
length - they only go in steps, they never record lengths like 83mm,
87mm etc - only 5mm or 10mm stages.

David

[David J Taylor]

I have taken tripod-mounted ISO 100 shots with all three cams (A900,
D3X and 5DMkII) using the same subject from a different angle, and
will be posting these over the weekend. Problem - all three cams have
very different true EI sensitivities, and I should have
flash-bracketed with the studio flash in 1/3 steps. I did not,
because until the tests were completed, I had no way of telling that
the Canon would be a full f-stop less sensitive at ISO 100 than the
Nikon (the Sony is in between).

Thanks, David, I appreciate the problems you had a little better.

For this ISO 1600 test, camera shake was part of the test.

I had not realized that.

I made
three exposures with each camera, taking care to be steady; 1/60th at
100mm-ish on a close up is about 1-2 stops past the 'safe'
recommendation (or 2-3 stops by the standards of pixel peeping). Both
the Sony SSS and Nikon VR systems claim to cover this; the Sony did,
perfectly each shot; the Nikon produced two shots with slight double
imaging visible from the VR, and one still not as sharp as the Sony -
as you can tell - which was the shot selected.

I keep wanting a really good independent test for VR, but ideally one with
a preset vibration input (like those 7-post jigs used for racing cars....)
rather than a human input which with tiredness etc. and may also depend on
how the camera is gripped.

Since you would not often use a tripod with ISO 1600, this was a
hand-held tungsten light test deliberately set up that way. The light
level also approximates typical stage/theatre/stadium values.

David

I might ask whether under those taking conditions you /need/ a 24MP
camera....but perhaps I should not ask!

I do appreciate the chance to see the images.

Cheers,
David

[David Kilpatrick]

First off, I'd like to see shots from a tripod with mirror up and in raw
converted with the software of Sony and Nikon. Then see with which you can
get the best result.

Then it's a question of which camera is easiest to handle, has the most
possibilities or limitations, etc. I never shoot in JPEG out of the camera,
so for me personally it's a useless test. I think most serious shooters,
that pay so much for a camera, will shoot raw too, especially with a 24 MP
camera.

Of course I am doing many further tests. This was a test of ISO 1600
tungsten light hand-held with image stabilisation, in-camera JPEG. If
it's useless to you, so be it, but that just to you.

Having used the D3 for a decent spell before getting the A900, I can
tell you which is the easiest to handle - the Sony. But so is the D700.

I also shoot raw only (or raw+jpeg, though since starting with the A900,
I have been sticking to just raw to maximise card space and battery
life). It's not a useless test for my readerships for the tests being
done, since many shoot JPEG - but I've been plugging the benefits of raw
shooting for several years, and so do many speakers on the seminar
circuits involved, the number of dedicated JPEG shooters is falling.

David

[John Sheehy]

I might ask whether under those taking conditions you /need/ a 24MP
camera....but perhaps I should not ask!

Oversampling is a good thing, in many ways. The only real problem is the
size of the data. Oversampled images have finer noise, finer bayer
artifacts, less moire and aliasing of any type, etc, etc. They rotate,
perspective correct, etc, with less loss of detail. They are more ideal
sources for downsampling to arbitrary sizes.

Furthermore, if motion blur of the camera is in a straight line, you can
deconvolve the blur more effecitvely.

[David J Taylor]

I might ask whether under those taking conditions you /need/ a 24MP
camera....but perhaps I should not ask!

Oversampling is a good thing, in many ways. The only real problem is
the size of the data. Oversampled images have finer noise, finer
bayer artifacts, less moire and aliasing of any type, etc, etc. They
rotate, perspective correct, etc, with less loss of detail. They are
more ideal sources for downsampling to arbitrary sizes.

Furthermore, if motion blur of the camera is in a straight line, you
can deconvolve the blur more effectively.

John, "good thing" I could agree with for the reasons you state, but I
still don't think you "need" such a camera. Yes, the /lens/ should be the
limit, not sampling in the sensor....

Cheers,
David

[David J Taylor]

If you turn up your shutter speed suitably to compensate for the
higher resolution, they will also be good for fast opportunistic
snapping where you have to shoot wide in roughly the right direction
and hope you can crop down to a reasonable composition later.

It's often forgotten that a 24MP camera is one that you can crop off
half the image, or lose half of it heavy transformative
post-processing such as doing architectural perspective correction
lens tilt computationally instead of optically, and still end up with
a sharp high quality 12MP image.

Well, yes, but you are back to the smaller sensor noise performance in
that case. I take your point, though. I still won't be getting a D3X
upgrade!

Cheers,
David

[David Kilpatrick]

I am getting a pair of Sigma 28-70mm f2.i8 lenses in Nikon and Sony
fit next week, to make totally 'equal' comparisons. That is, if the
Sigmas are really equal...

Again, would prefer single focal lengths for such tests. The Sigma 180
f/3.5 would be a good Sony/Nikon/Canon comparing lens v. most of their
zooms (assuming it's still made in full frame).

We tried for various choices, but this was all they had available. It's
already asking a bit much to borrow two lenses this way, when the end
result is not a test of Sigma products - they can't be sold as new after
I've used them, they have to go out as demo B stock once returned.

I don't think a 180mm macro would really be ideal, it limits the
possible subjects for comparison.

The only way I can make a pure comparison is by mounting my macro
bellows with a 75mm Apo Rodagon (nice lens, came out of a scrapped
Leafscan 45). It does tend to produce a flare patch but maybe some
attention to baffles, blackening various mount parts and fitting a
bellows hood could remove that. That way, I can at least compare the
Sony and Nikon sensors directly with precisely the same optical unit.

David

[David J Taylor]

Yes, the /lens/ should be the limit, not sampling in the sensor....

They are inseparable. re: Kodak equation.

1/res-out^2 = 1/lens-res^2 + 1/sensor-res^2

IOW you can increase the resolution of one or the other but never
achieve the resolution of the worst of the pair.

My comment was in a response to the mention of "oversampling". What I'm
trying to say is that the sensor should not be the ultimate limiting
factor, but the lens should be. In other words, make the sensor so good
that its resolution almost ceases to enter into the equation. High-end
guys like yourself should be looking towards this.

On the other hand, I can get away with a smaller and cheaper lower
resolution lens simply because I don't seek as many pixels, and hence I
will be more limited by the sensor. I'm not aiming at lens-limited
quality, just something that's rather better than the 2MP display I view
with today - give a bit of margin for cropping etc.

Cheers,
David

[David J Taylor]

That's all fine and dandy and until somebody corrected me recently, I
thought the same thing. On putting the equations into a spreadsheet
and trying some realistic numbers you find that the amount of
increased lens resolution over an existing high quality lens to get
even close to sensor resolution is unrealistically high (1/x tends to
infinitely fine resolution doesn't it?).

Flip side applies as well.

The trouble is, though, that people a single figure for "resolution",
whereas the shape of the MTF curve matters.

Well, in a way you are. As long as you stick to APS-C sensors at high
'pix counts, then you're even more affected than a FF at twice the
number of pixels. And you can only improve on that with better
lenses.

You say a 2 Mpix display is enough. Don't you print?

Very rarely, and then probably not more than A4 size (297 x 210mm). In
the worst case, with no cropping or processing, the display MTF completely
dominates my system, and allows me more freedom with lens and sensor
choice. Providing I have a good MTF up to the 2MP level, and sufficiently
good anti-aliasing so that artefacts aren't generated, I'm a happy bunny
with a cheaper 6-10MP DSLR and a cheaper and lighter DX-format lens.

I did enough black-and-white printing and some colour processing in the
past. My 35mm photos were mostly slides, though, Kodachrome and
Ektachrome. I'm much happier to see my images on a TV or computer screen
today. If I want big prints made I'll probably get them done commercially
by a firm like Photobox or one of the print shops in town.

Cheers,
David