By Carl Weese
Link to Amazon U.S. K20D page
Link to Amazon U.K.
K20D page
We're about to look at first impressions of the new sensor in the K20D, but first, the result of some serendipitous testing. Monday morning I decided to repeat two tests (white clipping point and ISO/noise) from last week, this time using both K10D and K20D cameras one right after the other so the files can be compared side-by-side. For ISO I set the tripod about seven feet from a cluttered bookcase. Lots of detail, many bright whites and dark shadows to show off noise. Without thinking about it, I used my 21mm DA Limited, letting the cameras handle focus while I fiddled with the ISO settings. I've had the K20D less than a week and already forgot about "the focus issue." The 21mm has been diagnosed by the K20D as needing a +6 custom AF focus point adjustment. The files from the K20D were in perfect focus, but the set from the K10D was off by a country mile. So bad that I couldn't use them to evaluate noise at 100% view.
This presented an opportunity. The K20D has given my 40mm DA Limited a clean bill of health: perfect AF calibration. So I repeated the series using the longer lens. Both cameras nailed the focus. A lovely, if accidental, confirmation of the accuracy and usefulness of the AF calibration feature. I'd also been wondering to what extent a calibration derived from tests of a small target at just a couple feet would hold at longer distances. This showed it holding perfectly at seven feet, and real-world pictures I've been making the past few days show it tracking fine into the middle and far distance. Here's what a simple focus test target looks like, as seen by the camera during the test:
Angle of approach should be 45 degrees. Make the AF focus on the dark line in center, then, at high magnification, see if the sharpest plane lies in front or back. (This is Tim Jackson's focus test chart.)
The sensor
Now on to that new CMOS sensor. The pixel horsepower race being what it is, of course the first point is the pixel count. At 14.6 MP it's not only a 46% increase over the K10D, it's also out front of the price-class competition, at least for the moment. So my first impression is, these files are big! Each RAW capture takes up 23.2 megs of storage space, so an 8 gig SD card fills up with 334 captures. 165 shots fill up an archive DVD. A 16-bit file opened from RAW into Photoshop weighs in at 83 megs.
The main reason to seek more pixels is to make bigger prints with less up-interpolation of the data. At 360 ppi (my standard since it's native for the Epson pro-stylus printer I use) the print has grown from the K10D's 7.2 x 10.75 inches non-interpolated or native file to 8.62 x 12.98 inches. Probably more important, the amount of up-interpolation needed to make a print 22 inches wide has dropped from 204.5% down to 169.5%. I have not made prints yet, because I will first have to test to find out what combination of interpolation and sharpening works best for these files. The effect of the higher pixel count is plain to see on screen, though. I have not worked with a CMOS sensor before and will have to feel my way into getting the character of this thing, but my first impressions are favorable.

From Town Plot, Waterbury, Connecticut
Sunday till mid-day there was luminous light under slowly clearing clouds. This capture, made on auto exposure with no compensation, holds the full range of values with beautiful tonal transitions right on the defaults in ACR.
In particular, from my first pictures of real subjects, I get a sense of smoothness, of more seamless tonal transitions throughout the range, while the general feel of the color is similar to the K10D. At the same time, detail looks very "optical" to me. That is, edges and textures look the way I expect a lens to render them, not as though they've been artificially generated by a computer. The terms "plasticky" and "film-like" get tossed around a bit—so far my impression is the K20D results are in the latter camp.
The main concerns when engineers pack almost half again as many pixels onto the same size sensor are a potential loss of dynamic range and possible increase in noise. For dynamic range, I'm most concerned to know how far up into the highlights I can go before losing data to clipping. To test this I place a smooth piece of card stock in perfectly even light and aim the camera on tripod at it. I make a bracket around neutral gray, based on the camera's spot meter (good time to check whether the other meter modes agree reasonably). Then starting at 2 stops over-exposure (Zone VII to classicists) I move the exposure up 1/3 of a stop at a time. Download and open the set of exposures as a group in ACR and check densities with eyedropper and histogram. The card exposed as middle gray came in at level 128 for the K10D, 119 for the K20D, a difference of something like 1/8th or 1/6th of a stop. Both cameras held full detail, no clipping, with space between the sharp peak of narrow histogram and the right wall, up to 2.3 stops above middle. Each showed a touch of clipping at 2.7: the K10D file came back with a Recovery setting of 4. The K20D needed a Recovery of just 2. At three stops up, the K10D recovered with a setting of 10, the K20D with 6. At 3.3 stops up, the K10D couldn't pull back from clipping with any amount of recovery, while the K20D came clear at 16—not an extreme amount of Recovery. At +3.7 it blew through the right wall of the histogram.
So, even with nearly 50% more pixels, highlight retention is slightly improved. In addition, my real world shots are indicating that this test may underestimate the improvement. In the test, not only does the new camera reach a little higher, but it holds more data just before the cutoff. The older camera sort of slides into the clipping above +2.3 stops. This conforms to my impression that the highlights in real pictures are richer, improved more than the white card test indicates.
Noise
Now for the noise tests, second go-round. The K10D's noise performance up to ISO 400 is good enough that in walkabout shooting I make a practice of using the camera's auto ISO feature, limited to a range of 100-400. I've never had a picture ruined by excessive noise because the light demanded 400. Up to 400, I also don't see a real difference between the K10D and K20D—except that a 100% view of the new camera's files is more magnified on screen because of the higher pixel count. At 800, the judgment gets a little tricky, which calls for a digression.
The K20D has a new menu item that lets the user select one of four levels of high ISO noise reduction: Off, Weakest, Weak, and Strong. I shot a series at ISO 100 through 3200 using RAW+best quality JPEG. As near as I can tell by pixel peeping at 100%, the NR is entirely done by the software that makes the in-camera JPEG files, not hard-wired into the sensor. The differences are plain to see in the JPEGs, but I can see no change at all to the RAW files viewed in ACR—they all look like the JPEGs done at "Off." This is fine with me, because to me the JPEGs with NR look worse, not better. Any reduction in noise is more than offset by smearing of detail, and of the noise itself.
Both at 800 and 1600, the noise in the K20D RAW files viewed at 100% in ACR looks better to me than the older camera, though at 800 I would not say there is exactly "less" of it. When the two are compared, the K20D noise looks crisp and clean, while the K10D files at high ISO look, in comparison, as though they've had some smeary noise reduction done to them, even viewed in ACR. In the old days, there were films that had smeary, mushy grain and films (famously 35mm Tri-X and T-Max P3200) that had beautiful, tight, clean-edged grain. Good grain that people came to love—imagine Salgado's pictures without that grain! At least on screen, the K20D looks like it may have "good noise" at high ISO. The new camera's 800 files look better, and the 1600 files look a whole lot better, than the older camera. 3200 looks really coarse. These are first impressions that await actual print tests for confirmation, but I'm looking forward to printing some good pictures made in rotten light at 1600. A little bit of nice tight grit may be just the thing for "available darkness" shooting.
Liquor Store, Union City, Connecticut
Another shot from Sunday with a long range of tones held beautifully with just a little adjustment in ACR.
Next time, more on color and tone, and more on new features.
_______________________
Previously:
Pentax K20D Report Part I: Focus
_______________________
Carl
Carl's website
Working Pictures
WPII