CS4 NOT capable of sharp displays at all zoom levels

Question

I must have been asleep, until now, and missed the significance and importance of what follows.
In post #11 here:
http://forums.adobe.com/thread/375478?tstart=30
on 19 March 2009 Chris Cox (Adobe Photoshop Engineer - his title on the old forums) said this, in a discussion regarding sharpness in CS4:
"You can't have perfectly sharp images at all zoom levels.". Unfortunately, my experience with CS4 since its release late last year has repeatedly confirmed the correctness of this statement.
What makes this statement so disturbing is that it contradicts an overwhelming amount of the pre- and post-release promotional advertising of CS4 by Adobe, to the effect that the OpenGL features of CS4 enable it to display sharp images at all zoom levels and magnifications. What is surprising is that this assertion has been picked up and regurgitated in commentary by other, sometimes highly experienced, Ps users (some unconnected with, but also some directly connected with, Adobe). I relied upon these representations when making my decision to purchase the upgrade from CS3 to CS4. In fact, they were my principal reason for upgrading. Without them, I would not have upgraded. Set out in numbered paragraphs 1 to 6 below is a small selection only of this material.  
1. Watch the video "Photoshop CS4: Buy or Die" by Deke McClelland (inducted into the Photoshop Hall of Fame, according to his bio) on the new features of CS4 in a pre-release commentary to be found here:
http://fyi.oreilly.com/2008/09/new-dekepod-deke-mcclelland-on.html
Notice what he says about zooming with Open GL: "every zoom level is a bicubically rendered thing of beauty". That, when viewed with the zooming demonstrated, can only be meant to convey that your image will be "sharp" at all zoom levels. I'm sure he believes it too - Deke is someone who is noted for his outspoken criticism of Photoshop when he believes it to be deserved. It would seem that he must not have experimented and tested to the extent that others posting in this forum have done so.
2. Here's another Adobe TV video from Deke McClelland:
http://tv.adobe.com/#vi+f1584v1021
In this video Deke discusses the "super smooth" and "very smooth" zooming of CS4 at all zoom levels achieved through the use of OpenGL. From the context of his comments about zooming to odd zoom levels like 33.33% and 52.37%, it is beyond doubt that Deke's use of the word "smooth" is intended to convey "sharp". At the conclusion of his discussion on this topic he says that, as a result of CS4's "smooth and accurate" as distinct from "choppy" (quoted words are his) rendering of images at odd zoom levels (example given in this instance was 46.67%), "I can actually soft proof sharpening as it will render for my output device".
3. In an article by Philip Andrews at photoshopsupport.com entitled 'What's New In Adobe Photoshop CS4 - Photoshop 11 - An overview of all the new features in Adobe Photoshop CS4',
see: http://www.photoshopsupport.com/photoshop-cs4/what-is-new-in-photoshop-cs4.html
under the heading 'GPU powered display', this text appears :
"Smooth Accurate Pan and Zoom functions – Unlike previous versions where certain magnification values produced less than optimal previews on screen, CS4 always presents your image crisply and accurately. Yes, this is irrespective of zoom and rotation settings and available right up to pixel level (3200%)." Now, it would be a brave soul indeed who might try to argue that "crisply and accurately" means anything other than "sharply", and certainly, not even by the wildest stretch of the imagination, could it be taken to mean "slightly blurry but smooth" - to use the further words of Chris Cox also contained in his post #11 mentioned in the initial link at the beginning of this post.
4. PhotoshopCAFE has several videos on the new features of CS4. One by Chris Smith here:
http://www.photoshopcafe.com/cs4/vid/CS4Video.htm
is entitled 'GPU Viewing Options". In it, Chris says, whilst demonstrating zooming an image of a guitar: "as I zoom out or as I zoom in, notice that it looks sharp at any resolution. It used to be in Photoshop we had to be at 25, 50 , 75 (he's wrong about 75) % to get the nice sharp preview but now it shows in every magnification".
5. Here's another statement about the sharpness of CS4 at odd zoom levels like 33.33%, but inferentially at all zoom levels. It occurs in an Adobe TV video (under the heading 'GPU Accererated Features', starting at 2 min 30 secs into the video) and is made by no less than Bryan O'Neil Hughes, Product Manager on the Photoshop team, found here:
http://tv.adobe.com/#vi+f1556v1686
After demonstrating zooming in and out of a bunch of documents on a desk, commenting about the type in the documents which is readily visible, he says : "everything is nice and clean and sharp".
6. Finally, consider the Ps CS4 pdf Help file itself (both the original released with 11.0 and the revised edition dated 30 March 2009 following upon the release of the 11.0.1 update). Under the heading 'Smoother panning and zooming' on page 5, it has this to say: "Gracefully navigate to any area of an image with smoother panning and zooming. Maintain clarity as you zoom to invididual pixels, and easily edit at the highest magnification with the new Pixel Grid." The use of the word "clarity" can only mean "sharpness" in this context. Additionally, the link towards the top of page 28 of the Help file (topic of Rotate View Tool) takes you to yet another video by Deke McClelland. Remember, this is Adobe itself telling you to watch this video. 5 minutes and 40 seconds into the video he says: "Every single zoom level is fluid and smooth, meaning that Photoshop displays all pixels properly in all views which ensures more accurate still, video and 3D images as well as better painting, text and shapes.". Not much doubt that he is here talking about sharpness.
So, as you may have concluded, I'm pretty upset about this situation. I have participated in another forum (which raised the lack of sharp rendering by CS4 on several occasions) trying to work with Adobe to overcome what I initially thought may have been only a problem with my aging (but nevertheless, just-complying) system or outdated drivers. But that exercise did not result in any sharpness issue fix, nor was one incorporated in the 11.0.1 update to CS4. And in this forum, I now read that quite a few, perhaps even many, others, with systems whose specifications not only match but well and truly exceed the minimum system requirements for OpenGL compliance with CS4, also continue to experience sharpness problems. It's no surprise, of course, given the admission we now have from Chris Cox. It seems that CS4 is incapable of producing the sharp displays at all zoom levels it was alleged to achieve. Furthermore, it is now abundently clear that, with respect to the issue of sharpness, it is irrelevant whether or not your system meets the advertised minimum OpenGL specifications required for CS4, because the OpenGl features of CS4 simply cannot produce the goods. What makes this state of affairs even more galling is that, unlike CS3 and earlier releases of Photoshop, CS4 with OpenGL activated does not even always produce sharp displays at 12.5, 25, and 50% magnifications (as one example only, see posts #4 and #13 in the initial link at the beginning of this post). It is no answer to say, and it is ridiculous to suggest (as some have done in this forum), that one should turn off OpenGL if one wishes to emulate the sharp display of images formerly available.

Answer 1

Thanks, Andrew, for bringing this up.  I have seen comments and questions in different forums from several CS4 users who have had doubts about the new OpenGL display functionality and how it affects apparent sharpness at different zoom levels.  I think part of the interest/doubt has been created by the over-the-top hype that has been associated with the feature as you documented very well.
I have been curious about it myself and honestly I didn't notice it at first but then as I read people's comments I looked a little closer and there is indeed a difference at different zoom levels.  After studying the situation a bit, here are some preliminary conclusions (and I look forward to comments and corrections):
The "old", non-OpenGL way of display was using nearest-neighbor interpolation.
I am using observation to come to this conclusion, using comparison of images down-sampled with nearest-neighbor and comparing them to what I see in PS with OpenGL turned off.  They look similar, if not the same.
The "new", OpenGL way of display is using bilinear interpolation.
I am using observation as well as some inference: The PS OpenGL preferences have an option to "force" bilinear interpolation because some graphics cards need to be told to force the use of shaders to perform the required interpolation.  This infers that the interpolation is bilinear.
Nothing is truly "accurate" at less than 100%, regardless of the interpolation used.
Thomas Knoll, Jeff Schewe, and others have been telling us that for a long time, particularly as a reason for not showing sharpening at less than 100% in ACR (We still want it though ).  It is just the nature of the beast of re-sampling an image from discrete pixels to discrete pixels.
The "rule of thumb" commonly used for the "old", non-OpenGL display method to use 25%, 50%, etc. for "accurate" display was not really accurate.
Those zoom percentages just turned out to be less bad than some of the other percentages and provided a way to achieve a sort of standard for comparing things.  Example: "If my output sharpening looks like "this" at 50% then it will look close to "that" in the actual print.
The "new", OpenGL interpolation is certainly different and arguably better than the old interpolation method.
This is mainly because the more sophisticated interpolation prevents drop-outs that occurred from the old nearest-neighbor approach (see my grid samples below).  With nearest-neighbor, certain details that fall into "bad" areas of the interpolated image will be eliminated.  With bilinear, those details will still be visible but with less sharpness than other details.  Accuracy with both the nearest-neighbor and bilinear interpolations will vary with zoom percentage and where the detail falls within the image.
Since the OpenGL interpolation is different, users may need to develop new "rules of thumb" for zoom percentages they prefer when making certain judgements about an image (sharpening, for example).
Note that anything below 100% is still not "accurate", just as it was not "accurate" before.
As Andrew pointed out, the hype around the new OpenGL bilinear interpolation went a little overboard in a few cases and has probably led to some incorrect expectations from users.
The reason that some users seem to notice the sharpness differences with different zooms using OpenGL and some do not (or are not bothered by it) I believe is related to the different ways that users are accustomed to using Photoshop and the resolution/size of their monitors.
Those people who regularly work with images with fine details (pine tree needles, for example) and/or fine/extreme levels of sharpening are going to see the differences more than people who don't.  To some extent, I see this similar to people who battle with moire: they are going to have this problem more frequently if they regularly shoot screen doors and people in fine-lined shirts.   Resolution of the monitor used may also be a factor.  The size of the monitor in itself is not a factor directly but it may influence how the user uses the zoom and that may in turn have an impact on whether they notice the difference in sharpness or not.  CRT vs LCD may also play a role in noticeability.
The notion that the new OpenGL/bilinear interpolation is sharp except at integer zoom percentages is incorrect.
I mention this because I have seen at last one thread implying this and an Adobe employee participated who seemed to back it up.  I do not believe this is correct.  There are some integer zoom percentages that will appear less sharp than others.  It doesn't have anything to do with integers - it has to do with the interaction of the interpolation, the size of the detail, and how that detail falls into the new, interpolated pixel grid.
Overall conclusion:
The bilinear interpolation used in the new OpenGL display is better than the old, non-OpenGL nearest-neighbor method but it is not perfect.  I suspect actually, that there is no "perfect" way of "accurately" producing discrete pixels at less than 100%.  It is just a matter of using more sophisticated interpolation techniques as computer processing power allows and adapting higher-resolution displays as that technology allows.  When I think about it, that appears to be just what Adobe is doing.
Some sample comparisons:
I am attaching some sample comparisons of nearest-neighbor and bilinear interpolation.  One is of a simple grid made up of 1 pixel wide lines.  The other is of an image of a squirrel.  You might find them interesting.  In particular, check out the following:
Make sure you are viewing the Jpegs at 100%, otherwise you are applying interpolation onto interpolation.
Notice how in the grid, a 50% down-sample using nearest-neighbor produces no grid at all!
Notice how the 66.67% drops out some lines altogether in the nearest-neighbor version and these same lines appear less sharp than others in the bilinear version.
Notice how nearest-neighbor favors sharp edges.  It isn't accurate but it's sharp.
On the squirrel image, note how the image is generally more consistent between zooms for the bilinear versions.  There are differences in sharpness though at different zoom percentages for bilinear, though.  I just didn't include enough samples to show that clearly here.  You can see this yourself by comparing results of zooms a few percentages apart.
Well, I hope that was somewhat helpful.  Comments and corrections are welcomed.