Type on the Panther Screen

Disclaimer:
This is not a platform battle. Since the early days my opinion has been that both major OSes suck to high heaven.

--

Allow me to be blunt: you'd have to be insane to think MacOS font rendering is great.

Please observe this graphic:
http://www.themicrofoundry.com/other/aa-compare.gif

The top and bottom two rows are from that article. Reminder: the top two use sub-pixel rendering, the bottom two don't.

The middle two rows are from Windows XP-Home: the top is plain ClearType, the bottom is a tweaked Times to render anti-aliased at 12 PPEM without using sub-pixeling.

I used Times because that's the closest thing to Hoefler Text that I have. It's slightly larger on the body, but also narrower, so it's about the same in terms of space usage. As for hinting, Times has decent -but not sublime- hinting, while it doesn't matter what Hoefler Text has because Panther ignores it.

And let's not even get into italics.

--

Please let's all admit Apple has a long way to go, otherwise we'll be stuck with this crap for a while. It's bad enough that MacOS totally ignores embedded bitmaps, at least they should get to MS's level of automatic rendering from outlines so that fact doesn't hurt as much.

hhp

I suppose I'm so used to Jaguar that I marvelled at the improvements in Panther. Hoefler Text looks so good in Panther - especially the italic - that I'm using it as my default browser font. I'm using it on a Pismo LCD screen, which might look different from whatever you're using, Hrant. Versus Cleartype though, I think Cleartype still has an edge, no pun intended. But versus standard Windows fuzz, it's much nicer. And Times in Windows versus Hoefler in Panther is apples & oranges.

One thing you need to take into account is the difference in gamma settings. The standard gamma setting in Windows is significantly darker than the standard Mac gamma. Your Non-CT Hack looks very washed out and spindly on my Mac (flat panel) screen. When you see these screen shots on your Windows laptop, you are not seeing the same thing Mac users see.

Also, you cheated a little bit: Your caps and lowercase characters are both one pixel taller than Gruber's samples. A pixel can make a big difference at that size. It would be better to compare Times at the same pixel size rather than Times vs. Hoefler Text.

> especially the italic

I'd like to see it. Here's Windows:
http://www.themicrofoundry.com/other/tita_ct.gif

> standard Windows fuzz

What exactly is that?

> Times in Windows versus Hoefler
> in Panther is apples & oranges.

Come on, don't exagerate.

> difference in gamma settings.

True. On the other hand, gamma affects the midrange. It doesn't affect fuzz very much, and no amount of gamma difference is going to make that horizontal bar in the "H" on MacOS solid black.

> Your Non-CT Hack looks very washed out
> and spindly on my Mac (flat panel) screen.

Same here (although certainly less, because of the gamma thing). But it's something you're never "supposed" to see - it's one reason MS turns off anti-aliasing for "normal reading" range. The reason I put that hack in there is to show the quality of the Windows rendering algorithm, with no help from subpixels. If you used a slightly darker font with no hairlines (like in that "2"), it would look superb.

> you cheated a little bit

Well, I wouldn't call it cheating...
1) I don't have Hoefler Text for Windows.
2) Times is narrower, and that takes away form the room you have to play with too.

You don't have to have an exact equivalence to see the superiority of the Windows rendering.

But I would love to see that test done with Times on the Mac side. As much as Hoefler Text is a better text font than Times on paper, on screen it's a bad benchmark for two reasons:
1) It's small on the body, and the counters are too closed.
2) It's not references in HTML nearly as often.

(I wonder if Gruber used Hoefler Text exactly because he wanted to avoid a direct comparison with Windows...)

hhp

Hrant's complaining about AA, so it must be wednessday.

anyhow.
Did anybody notice the condensed version of Lucida Grande that kicks in at point sizes

I only get one day a week for that? :-)

Lucida Grande: Can you elaboarate? And I think you're missing a word up there.

--

So just to be clear: could somebody please provide Times screen grabs at 12 point from Panther? With and without subpixeling, if possible, and optionally italics too. Thanks.

--

And just to be explicit:
> standard Windows fuzz

There is no such thing.

Fonts that render badly in Windows will render badly in MacOS too. But Panther's "ceiling" is much lower.

hhp

"standard Windows fuzz" = the non-ClearType, pre-ClearType antialiasing. It's still an option in Display Properties. Most people use either None or ClearType.

John, when CT is turned off the MS core fonts are internally defined (GASP table) not to anti-alias at certain sizes no matter what settings are chosen. These sizes are the sizes we're talking about here. That's why I had to do that hack to show the merits of the Windows rendering algorithm (sans CT), in spite of the spindliness of Times.

Typically sizes between 9 and 17 PPEM (inclusive) are simply 1-bit (b&W), hence there can be no fuzz. Above 17 is grayscale, but thanks to the quality of the algorithm it's bearable upto about 21 PPEM, above which even I have no qualms.

hhp

You guys aren't sending me grabs of Times in Panther because you're embarassed. I would be too, but you won't get your favorite OS to improve its text rendering by pretending it's acceptable as it stands.

hhp

No, it's because we have jobs, Hrant. And I'm not the one trying to prove a point. I don't feel any sense of loyalty to any OS. (The things people substitute for religion these days...)

I have jobs too. One of them is to learn, objectively.

Since you're luckily free of platform fanaticism, please consider devoting 5 minutes of your free time to this, when you get a chance, at your leisure, no pressure. OK?

hhp

I haven't installed Panther yet. I plan to do so soon. When I do I'll post something if no one else has yet. (I tend not to do major system upgrades during crunch times, which is the case for me just now.)

Try these, Hrant. They're both from Panther: the first from TextEdit (subpixel), the second from Word vX (grey only).

Panther subpixel Times sample (TextEdit)
Panther grey Times sample (Word)

Each sample shows a line each of Times, Times Italic, Times New Roman, Times New Roman Italic all at 12pt and 100% zoom. As far as I can tell, TrueTypes are being used in all cases.

Note that you have to use an application's own subpixel "disabler" in order to render in pure grey. This may invoke an app's own renderer in some cases - certainly these are pretty ropey on my screen (e.g. '2' in top line, 'z' in bottom line). Panther, and Jaguar if I recall correctly, seem to discourage pure greyscale rendering: surprisingly, its "Standard - best for CRT" mode performs subpixel rendering. I wonder if there's a hack... Word also does some font substitution by default (if you simply paste in the rich text from TextEdit), which I think I avoided.

The images were prepared using OS X's screen grab utility (a load of control keys, plus '4'), then pasting into PhotoShop, then saving as PNG.

Thanks Laurence! (And thanks for the intention, Mark.)

I only have one Times on my system.
BTW, wouldn't one of yours be a PS font? It's interesting that in the subpixel the two are virtually the same (except for the TNR being slightly lighter), but in the "true gray" they're so different. Is hinting used for either mode? I thought not.

Anyway, to avoid a combinatorial explosion or something I decided just to compare the best/most_typical of each platform. I chose the subpixel TNR on Panther because from the italic "z" you can see that's the one closer to the Windows Times. I think...

I had to equalize the gammas*, and to be fair I applied the same shift to each sample (about 6% lighter/darker for MacOS/Windows). This seems to have introduced some greenishness in the Windows settings, but I'm not sure how else to do a fairer comparison (except to physically have the two platforms side-by-side).

* BTW, PNG files [can] include gamma information, so they display differently ("correctly") on different platforms. You don't think your files are doing that though, do you?

--

Am I failing to account for anything?

Anyway, here's what I got:
http://www.themicrofoundry.com/other/aa-compare2.gif
The top of each pair is Panther, the bottom Windows.

Discuss...

hhp

It surprises me that no one seems to mind the color fringing that ClearType and the OS X equivalent creates. It jumps out at me at all sizes, and not in a good way. I've even been noticing it in magazine screen shots lately. Ug-ly. I'm also amazed by how artificially bold all three levels of OS X subpixel antialiasing look on my (LCD) display compared to "Standard - Best for CRT," which looks by far the best to me.

Hrant--

The "pure grey" samples that Lawrence shows from Word are created by QuickDraw, the old Mac graphics engine, not Quartz. This is the way everything looked in pre-OS X Mac systems. Word, as a Carbon app, defaults to QuickDraw anti-aliasing when Quartz anit-aliasing is disabled (in Word).

The only way to get "pure grey" Quartz anti-aliasing is to use the "Standard - best for CRTs" setting. The other three settings use sub-pixel rendering.

Max--

I assume you are on a Mac. What kind of LCD are you using? If you are seeing strong color fringing, it may be that you are using a 3rd party LCD screen that has the RGB elements in the reverse order from Apple's displays. An easy way to test this is to take one of the screen snapshots (above) and rotate it 180 degrees in Photoshop (or whatever). If it looks better, then that's what your problem is.

This can happen with ClearType on Windows too, and there is a setting to fix it. I don't know if this is possible to fix with OS X and Quartz or not. It's certainly not in the System Preferences, but it may be possible to change the setting from the terminal. Other similar settings which are otherwise inaccessible (such as being able to turn anti-aliasing off completely) can be changed from the terminal, so it may be also.

Mark, thanks for pointing out that Word etc. use QuickDraw when you disable Quartz rendering. I intended to use "Standard - best for CRTs" to generate the grey samples for Hrant but was surprised to see they perform subpixel rendering too - that's why I didn't use it for those images I made. Have you managed to get pure grey from this setting?

Your comparisons, Hrant, are interesting. I can't choose a clear winner - in the Roman, Windows seems to win on consistent horizontal stems (most caps) but screws up over inconsistent lc character widths (p,z,y) and poorly located serifs (u,n,b,l,p,d). Reading-wise, I adjust to Panther's consistent flaw of poor horizontal stems in caps more easily than ClearType's unpredictably odd letter widths in the lowercase.

Conclusions I can make (and I'm running Panther at home, and Jaguar at work):

The old OS 9-style of anti-aliasing stinks at small sizes--especially on italic Times, which is exactly why I used to turn it off.

When I look at the screen shots prepared by Laurence and Hrant, it seems the Panther text on Hrant's image is much lighter, to the effect that the tops of the capital letters, which are faint but visible on Laurence's shot, almost completely disappear on Hrant's. I know he tried to compensate for gamma, but that makes no sense.

Another thing: Gruber's column was about Panther vs. Jaguar. We can all applaud Apple for improving the anti-aliasing in Panther, right? It *is* better.

(sidebar: I have noticed too that Lucida Grande looks different at small sizes in Panther. Lucida Grande 10 definitely looks narrow-ized, as Rob suggested. I think they bumped up the vertical size and made it narrower to keep the width the same but improve legibility.)

Either way, I think it's safe to conclude that ClearType has the edge in legibility with Times New Roman. The letters are sharper, more clearly defined, and less "artificially bold" (vis à vis Max's complaint). Look at the roman 'x' for example. ClearType is also, however, slightly larger, as evidenced by the longer line length. I'd be VERY curious to know which size more closely matches a print out at that point size. That should be important.

One of my major misgivings about Jaguar's anti-aliasing was that it actually changed the size of some text. Georgia, in particular, takes up much less space than it did without anti-aliasing.

But you know, this is pretty academic. Times New Roman has always looked pretty bad at small pixel sizes. Anti-aliasing makes it look more natural, and less like the tragedy of 1980s 9-pin dot-matrix printers, but it doesn't really make it easier to read. I don't think anyone can seriously argue that Times 12 with *no* anti-aliasing is harder to read, it's just that it's also dog's-breakfast ugly.

My REAL conclusion? Never use Times New Roman at 12 pixel size for anything you expect people to read on screen! And if you're editing a document in Word on a Mac with 12pt text, for heaven's sake, zoom in to 150%.

Lawrence--

In order to see the change you made in the anti-aliasing setting, you need to quit and resart any running applications, in this case Word.

Mark --

I have an Apple Cinema Display, so I doubt that's the case. I must just be more sensitive to color fringing than most people.

A couple of points.

As noted in Kevin Larson's talk at ATypI consistently 10% of people claim to prefer other rendering methods over ClearType - the so-called 'developer effect' - maybe Max is one of those.

This page may help him determine if his system is set up correctly - http://www.microsoft.com/typography/faq/faq15.htm - I'm not aware of an Apple equivalent page.

This may be off base but looking at this - http://www.apple.com/displays/acd23/ the display has a pretty coarse resolution. Quite a bit coarser than your average laptop display. The coarser the resolution the more color fringing you're going to see.

Secondly, as hinted in some of these postings the Windows XP ClearType implementation has some limitations, specifically around ensuring backward compatibility and in working with core fonts designed and hinted before ClearType was invented. I think the best current ClearType implementation can be found in Word's new reading pane view, and would encourage those interested in the field to take a look at it.

It's true that my Cinema Display is something like 86 ppi, which is admittedly low. However, I don't like the color fringing on my iBook, either, which is 106 ppi. Maybe it's some combination of good vision and viewing distance.
Anyway, "standard" antialiasing looks fine to me, and to throw in a platform wars angle, I've never seen type on a Windows machine that looked close to passable--either it's not antialiased at all, or it's antialiased and spaced horribly, or it's excessively sharp and overhinted (like hrant's samples), or whatever. This shows how subjective on-screen type assessment is, I guess, but I've never seen a screen full of decent type on Windows, and the googling for XP screenshots I just did confirms it.

Max,

Passable text on screen is certainly rare, although I think progress is being made. The number of people who think ClearType and its clones are a step in the wrong direction is small. I mean you've switched to Panther - what do you think? Are Apple making a mistake going down this path?

Anyhow the most passable text I've seen on any platform has been rendered in the Word 2003 reading pane. You need to look at this running natively - Googling won't do the job.

Max, the fringe-ing bothers me too. But I think that a) if you have a hi-res screen and b) if you work out your display settings (like gamma) carefully, it's worth trading some color fidelity for triple horizontal resolution, especially for italics.

> I've even been noticing it in magazine screen shots lately.

Interesting! Screen grabbing isn't what it used to be, I guess.

> artificially bold

That might be:
1) Your gamma setting.
2) What you're used to. My view is that screen fonts have been too light (in reading sizes) for a while now, and the "developer effect" makes it harder for us -but no so much laymen- to re-adjust. For example, on Windows a standard font at 12 point (which is actually 16 PPEM) has one-pixel stems: too light.

> The "pure grey" samples that Lawrence shows from Word are created by QuickDraw

Interesting. That's why it looked so crappy (which is why I avoided using it in the comparison).

BTW, John Butler and Brendan Young have emailed me more grabs, and I'll be passing those along soon.

> I know he tried to compensate for
> gamma, but that makes no sense.

Right, it's not possible to do a "perfect" comparison (except live). One way I can improve it though is by accounting for the fact that the eye sees each color channel differently*: when I tweak the gammas I can weight it differently for each channel. Does anybody I remember the proportions? Laurence once providing some such info somewhere, no?

* Green for example is seen much brighter, which I guess explains the extra green in the tweaked Windows sample.

I think the trick is to look beyond the gamma, beyond the relative weight difference, and try to judge the quality of the algorithm, irrespective of the font. For example, to me the Windows rendering is consistently more "firm" and less fuzzy (although it also seems to use more color), and I think that's crucial.

> We can all applaud Apple for improving the anti-aliasing in Panther, right?

Certainly. But it's really not much of a laurel to rest on, especially when Apple has such a "designer" prestige thing going. If you remain satisfied with Panther's text rendering, chances are less it will improve sooner. Remember, the Windows stuff I'm showing has been around for a while now, and the non-CT algorithm is nothing short of ancient.

> I don't think anyone can seriously argue that Times 12 with *no* anti-aliasing is harder to read

I do!

I even have some empirical data to back it up, and Kevin Larson probably has a closet full of evidence.

Times does suck as a screen face, but in a way that makes it a good benchmark! And if anybody sends me a MacOS Georgia waterfall, I'll do some Windows comparisons of that too.

> it's excessively sharp and overhinted

Can you elaborate?

hhp

OK, I've taken a very close look at all the samples (thanks all), and here's two conclusions:
1) For the subpixel stuff Laurence's grabs are as good as it gets, even though they're especially dark. Brendan's sample was very light for some reason, while John's was great, except he got creative with the pangram ("humped" indeed :-) so I didn't get the A-H caps nor a "j" nor an "s". But from the top bar of the "T" you could tell it had the same problem of... firmness? :-) John also noticed some troubling differences between different versions (like CE) of Times.
2) On MacOS, TNR seem to render slightly better than Times, for some reason.

I thought I could do a better "normalized" comparison between MacOS and Windows by accounting for the eye's differing sensitivity to each of the three color channels (30% for Red, 59% for Green and only 11% for Blue) but stuff I've tried so far has actually backfired...

BTW, Simon, exactly what's the advantage of the Word 2003 "reading pane". Basically better CT rendering? Using hints or something?

hhp

I've asked the CT team for a high level explanation of the differences, but some of this isn't too difficult to work out when you consider the compromises that had to made to make sure that the CT text behaved in exactly the same way as the non-CT text.

The CT group couldn't help me here. All I can say is that the reading mode implementation of ClearType represents another year or more of research, as well as freedom from the backward compatibility restraints of the Windows XP implementation.

> if anybody sends me a MacOS
> Georgia waterfall, I'll do
> some Windows comparisons

Luke Andrews was kind enough to provide those to me the other day, but as he and I both found out soon enough, the rendering of Georgia varies between the two platforms in pretty much every single way, including vertical proportions! So no decent comparison was possible. Strange...

hhp

You will want to replace Panther's Georgia with with the latest georgia*.ttf files from Office 2003. There are far more characters in the latest Georgia. Apple's version is several years old. The addition of more glyphs will increase the leading (if at least one of those additional glyphs has a higher ascent or lower descent than the rest.)

I haven't tried this myself, so I can't guarantee it won't kill your machine. Back up the old Georgias (just rename them Georgia.old or something) and be prepared to boot in single-user mode to replace them if something goes wrong.

An interesting (if somewhat skewed) page:
http://www.xvsxp.com/fonts/

(Thanks to Chris Lowery for the find.)

hhp