I've owned two Radeon HD's, a 4670 and a 5670. The stability of the open source driver is good and its performance is good in most cases. Their proprietary driver is totally broken and dangerous to use. If you're considering buying an AMD card to try to run games and professional software with, you should take the following warnings.
AMD doesn't listen to or respond to most of their bug reports http://ati.cchtml.com/buglist.cgi?qu...duct=&content= for their proprietary driver, including one by me. http://ati.cchtml.com/show_bug.cgi?id=463.
Most of their driver releases are incredibly unstable and WILL (there is NO IF, only WHEN) cause data loss, kernel panics, X.org Server collapses, etc.
Sometimes a simple thing like putting the monitor in standby mode will make the system unable to wake up. (Which was the subject of my bug report.), this has been ongoing for at least a year, I only decided to file a bug report because this way I have something to point to when they say "where is the bug report?", but it should be obvious to most people that this happens. It's kind of hard to not notice your system going to sleep and not waking up, and having to hold down the power button.
In their Catalyst 12.4 beta, sometimes simply creating an OpenGL context will seems to panic the kernel, though this seems to be a race issue and of course it freezes the X server, so you see whatever the X server drew to the screen before the system died, not the actual kernel panic.
Last year, for two months (11.1 and 11.2), sometimes viewing a PNG with an alpha channel would cause the X server to collapse. Between 11.10 and 12.2, using an application which used X Video would cause the server to crash.
AMD's proprietary driver does not EVER support the current stable X server. If it does, it will be extremely error-prone, even for Catalyst, until another stable X server is out that Catalyst will again take months to support. Their kernel support is sometimes a release or two behind. Sometimes just fixing a bug in a current STABLE KERNEL SERIES will cause their module to not build (like it hasn't on x86-32 since 3.2.12 came out).
AMD has a number of technical problems with their Catalyst driver that are caused by it being developed outside of the kernel and outside of X.org with basically no quality control process. AMD's John Bridgman claims they have one, but I have yet to see a driver that looks like it has been through any kind of QA. It seems instead that they simply shit another one out every month and say it's a new version when it usually has all the same problems it ever did and maybe more.
The driver bundle is obese. The download size is well over 100 MB COMPRESSED, that's larger than an entire Linux kernel source package and all of X.org, combined. Even at this clownish size, it only manages to support about half of the Radeons that are out there. Anything pre-Radeon HD has no official support from AMD.
I understand that the reasons for some of this are:
(1) As a hardware company, they don't want to support cards that are more than a couple years old.
(2) They suck and are unable to compete with Nvidia. Which supports every card they've ever released and usually provide betas that support THE NEXT KERNEL AND X SERVER!!!!!
Why can't AMD do this? They wait for Microsoft to sign the freaking Windows driver and they hold that month's Catalyst Linux driver bundle hostage until Microsoft gets around to certifying the Windows counterpart. They release betas for Windows users without the Microsoft certification all the time, and they even have a Twitter account where their employee's advice is frequently to just use the uncertified betas to fix whatever their current release is doing wrong on Windows.
Just stop. Kill your evil driver or fix it please. It's getting to be a broken record on IRC in any distribution's chat room of "I installed Catalyst and now my system is fucking broken". It comes up, predictably, about a dozen times a day that I see, in one particular room.
And Bridgman, please stop ignoring me when I mention a bug and stop telling me you have a QA process when you clearly do not and it's anyone's guess what programs it will savagely maul that month because you can't make it work right.
Thanks. And if anyone is looking to buy AMD's hardware, take this as a warning that it's very hard to get it to do anything terribly useful, and you might as well just get a cheap one if you possibly can, because you're not going to get serious 3d performance or OpenCL support out of the open source stuff anytime soon.