How does the D3 achieve such high ISO?

Started Feb 17, 2008 | Discussions thread
Marianne Oelund Veteran Member • Posts: 7,788
Let's normalize things to the same ISO

PIXmantra wrote:

Marianne Oelund wrote:

I own both cameras, and have tested both extensively. I do not rely
on other's test results, which are too easy to misinterpret.

...But the same thing could be said about your tests or conclusions
(which I do observe, BTW).

Indeed. However, I would openly invite any whom are interested and locally available, to perform tests with my cameras.

Fact #1: Quantum efficiency is significantly higher on the D3,
yielding greatly reduced shot noise.

...It is higher but not as efficiently coupled with its actual Full
Well capacity. As a result, you can achieve higher dynamic range with
the 1D3, and a better or more-balanced distribution of sensitivity
and performance that includes ISO50.

What you are actually doing now (and Roger Clark fell into the same error in compiling his Figure 4), is to compare the Mk III at ISO 50, with the D3 at ISO 200! All that this proves, is that the D3 DR at ISO 200 falls short of the Mk III at ISO 50 - should anyone be surprised? Ask yourself: How does the Mk III at ISO 200 compare to itself at ISO 50?
Let's put the comparison on the same basis: ISO 200.
D3 full-scale e- count = 65,600 (Roger's number)
D3 read noise = 6 e- (my number, more conservative than Roger's)
Mk III full-scale e- count = 22,000
Mk III read noise = 4 e-
D3 sensor DR = 65,600/6 = 11,000 = 13.4 stops
Mk III sensor DR = 22,000/4 = 5,500 = 12.4 stops

The most important point here, though, is that huge 3:1 ratio of electron counts, which holds as we ascend the ISO scale. We'll see below, how significant this is to noise levels.

Fact #2: Read noise levels are nearly the same on the two cameras,
across the ISO range.

Well, the 1D3 seems to have LOWER levels of read-out noise, indeed.
This is exactly what allows it to attain to a slightly larger dynamic
range, at ISO100, than anything the D3 can attain at any ISO... And
that is with substantially SMALLER, 7.2 micron sensels! :-)))

Again, we must use the same basis (the D3 does not actually have ISO 100). Expressing read noise in absolute electrons fails to correctly compare the noise level at the A/D converter outputs, when the full-scale electron counts are so different (3:1 in favor of D3). What is the noise expressed in DN, which is what we actually see in the image?
For example, at a moderate sensitivity of ISO 800:
D3 full-scale e- count = 16,400
D3 read noise = 6 e-
Mk III full-scale e- count = 5,500
Mk III read noise = 4 e- (using an optimistic value; actual is a bit higher)
D3 read noise = 6/16,400 * 16,383 DN = 6 DN
Mk III read noise = 4/5,500 * 16,383 DN = 12 DN

Thus the D3's read noise component at the A/D converter output is only half that of the Mk III (note my "Fact #2" statement previously was based on Roger's older data for the Mk III of 2.1 e-, which he has since updated).

The result: The D3 is superior across the board, in its RAW output,
and its advantages regarding JPEG output are even greater.

The D3 shows its 0.5EV advantage in the ISO2000-ISO4000 range... but
it actually seems to offer DISADVANTAGES at any other lower step of
the scale. That is thanks to 1D3 usable (and superior) ISO50-ISO200
performance and dynamic range, not to mention superior per-pixel
quality and sharpness, most notably in .JPGs images (the 1D3 are
clearly more detailed and sharp, with less sharpening effort), all
the way up to ISO3200.

As already addressed above, the Mk III is inferior at ISO 200. Comparing lower ISO values is not possible, as the D3 doesn't offer them.

The Mk III has very aggressive JPEG sharpening, but this acts to its detriment in regard to noise, if one is not careful. I find that to achieve comparable JPEG sharpness, the Mk III sharpening must be set to "0", with the D3 at "3" or "4". With those settings, detail capture is equivalent, yet the D3 still shows lower noise levels, even with NR turned off. With the D3's NR turned on, it's simply no contest.

...Well, R. Clark's finding do not seem to be very aligned to yours
(notice the Perceived Image Quality model that brings noise and
sharpness together):

I'm not convinced of the accuracy of Roger's AIQ model, but putting my doubts aside for a moment, he gives the D3 a rating of 69, to the Mk III's 57. Is not the higher value the more desirable one?

At the end, the result boils down to:

From ISO 200 and up, the D3 significantly outperforms the Mk III, in both DR and noise levels, if one is careful to use data for corresponding ISO settings as the basis for comparison. To do otherwise results in a serious miscompare.

Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow