Secondly, if it's good enough for most users, why all the discussion? Why all the complaining, fudding, and carrying on in every forum, blog, bug report, etc, etc. Certainly this has to be one of the most vocal minorities ever.
Third, if things are as you say, "90%" and "typical", how can you be sure that people are happy with the open drives and haven't just given in and accepted that FLOSS will never offer the extras such as gaming and just turn to something else (ps, wii, xbox, windows box, etc). How much feedback bias is going on where fact that people can't influences that people don't.
Lets say tomorrow, that someone waves a magic wand and all the binary drivers were open, pattens were abolished, Epic came through with UT3, Steam etc... all happy rainbows and shit. Are you saying that the typical 90% won't give a shit because Supertux, KPatience, and Firefox have all been "fine" for some time?
(Oh wait scratch Firefox, they're not fine with most FLOSS drivers either )
How do you suppose Linux developers should provide an obsolete ABI and at the same time completely rebuild the graphics stack in a timely manner? And how does that work economically? What's the incentive for these developers to expend these huge resources to keep the ABI stable? An ABI that these developers don't even use?
Just buy Nvidia guys. They're power hungry, overpriced graphics cards but they work a billion times better than anything ATI ever made.
Right now, Nvidia barely works under Linux, and this is thanks to reverse-engineering. Nvidia does not offer me what I need, so I have turned to a company which does (AMD).
Nvidia makes good cards for running Windows, though. If that's what you want.
RE: icculus: "dogmatic attitudes towards software freedom"
Dogmatic attitudes towards software freedom? REALLY? This is the differentiating factor of GNU/Linux. Not that it's fast; not that it's developed efficiently "in the open"; not that it's more secure than Windows. People trying to pervert the vision of GNU/Linux into just another OS are the ones who are really dangerous.
Even pure "open source" people agree with me here, throwing freedom out the window: their thesis is that software is better (more efficient, less buggy, less expensive to develop, more featureful) if it is open source, than if it is proprietary. So they would argue that the binary drivers aren't as good as the open drivers just in principle. That they aren't (yet) better in fact is a matter of the open drivers playing catchup.
Consider the functions f(x) = x + 60 and g(x) = x^2 - 12. If you look at x = 3, of course f(x) is going to be better. But if you take it out to x = 16, g(x) has easily won the day, and will continue to diverge, growing faster and faster. That's the kind of behavior we have right now with the open drivers.
However, what you (icculus) seem to not understand, is that the very method by which the open drivers will evolve (if they evolve at all) involves, crucially, widespread testing. If you don't push the open drivers out in distros, your testing pool is going to be something around the typical size of #radeon on freenode. This is insufficient for two reasons.
First, it's obvious that only 100 or so people won't have all possible hardware combinations and ASICs. Some (untested) cards may not work at all because they have a different pinout, or strange/new display connectors (such as displayport) that are not yet supported. Making the developers aware of this is paramount to developing a good driver.
But more importantly, 100 people won't be running the breadth of applications that a few hundred thousand are. You of all people should know that the GL specification (and implementations) are extremely finnicky, and very likely to produce incorrect behavior at the slightest misstep, either on the driver side or the app side. Building in support for the quirks and nonstandard behavior of hundreds of existing apps, let alone new ones, is no minor task.
All that said, I agree 100% with your assessment that, in a pipe dream world, S3TC and floating point textures would be a non-issue that we could just plow through, and the vendors would share the code for fglrx and the nvidia binary with the world.
But banking on a pipe dream and ignoring your roots (i.e. the reason GNU/Linux has gotten as far as it has, i.e. dogmatic principles regarding free and open source software) is not going to endear you to anyone.
Dogmatic attitudes about software freedom are THE defining difference between Linux/GNU and something like Solaris or AIX.
There's a huge difference between Freedom and Open.
I enjoy having an Open core OS.
I also enjoy using software that actually works, and will give up a little bit of Open (or even a lot of it, in some cases) for higher quality software.
This is very different than the FSF ideals where _everything_ must be Free at all times, period, the end.
The problems with the Open drivers have nothing to do with testing. The problem is not "they don't work on some wild piece of hardware."However, what you (icculus) seem to not understand, is that the very method by which the open drivers will evolve (if they evolve at all) involves, crucially, widespread testing.
The problem is "they are missing vast swathes of functionality critical to modern graphics engines, and the functionality they do have runs at a fraction of the speed of the proprietary drivers."
The average Linux user may not care much (Linux users are by necessity generally not the game playing type), but the average computer user in general (some ~70% of which play games) is going to really care about this. If they can't run their silly little $10 bargin bin game because it just happened to some simple graphical effect that requires a proprietary driver, or it happens to hit one of the many slow paths in the current Mesa drivers, they're going to (rightly) assume the Linux desktop is a piece of crap and go back to Windows.
I've seen this happen time after time. Seriously. Every single last Linux desktop install I have ever known outside of my own has been replaced with Windows because users have decided that $100-$200 is a small price to pay for "just fucking works." Hell, even I'm getting damn close to going that route after 10 years of Linux desktop use, especially with how utterly screwed up the desktop is on all the new post-GNOME-3.0 distro releases.