05-29-2009, 07:01 AM
Personally I think that ATI is supporting open source better but closed drivers are still crap. Example is that thread: http://phoronix.com/forums/showthread.php?t=16644 . I got one nvidia card in my old computer and it works better than ATI card in my new computer, what the hell?
05-29-2009, 08:06 AM
The message is obviously biased and targeted at server hardware drivers, which generally are very thin layers between kernel and rather high-end hardware. The main reason in server space to use binary drivers is to hide hardware bugs, which might result in a bad PR. (I have programmed a number of such drivers, so this is a first hand account.)
Originally Posted by energyman
Most of the Greg's technical arguments boil down to the fact they kernel developers do not want to do it and most Linux server vendors do not need it.
N.B. And binary interface isn't only option BTW. I wonder who and why always drive the discussion in the direction, which is known to be a dead end.
3D hardware and their drivers are absolutely different beasts. The drivers are not thin - they are very thick and contain layers of logic. Also they interact with many kernel subsystems. 3D hardware has magnitudes more features, requires magnitudes more optimizations and magnitudes tighter integration with rest of the system - to deliver any sort of passable performance. There is inside of kernel no single driver subsystem which has comparable complexity.
Good luck trying to explain that to generic kernel developers.
I glad that wide Ubuntu adoption on desktop started raising the similar concerns with kernel developers being highly (and often blatantly open) server-hardware biased. It's not a secret and all are well aware that server companies roll lion share of Linux development. The problem is that improvements for desktop (and other areas) gets shuffled deep down the list of priorities, making it rather hard and unrewarding for willing people to contribute.
IOW, problem is not that there is nobody willing to improve desktop experience, problem is that getting improvements past the server-hardware bias is complicated enough to make people loose any will to do it.
05-29-2009, 01:40 PM
Sorry, just in case I wasn't clear, saying drivers are "crap" really doesn't say why you think they're crap. Was it just because of compiz? That was the only thing in the thread that had any substance to describing a problem. I would simply like to know why you think they're crap, and what makes you think nvidia's are better. If it's purely compiz, then say so, I just want to know if that's all.
Originally Posted by Mazur
05-29-2009, 03:44 PM
As I already said, or maybe not?, I like AMD as the company because they are creating really great piece of hardware but drivers for graphics card is 'crap'. Why it is 'crap' and why I think nvidia is better in this way? This is mainly because I am running Arch Linux and I cannot even get support for new kernel, those drivers was removed from 'Extra' repository to AUR because of bad support. AMD is good company because it supports open source, they are releasing new code etc. But it is too slow. Even performance on by old card is better than on my new card because of those drivers.
Originally Posted by mirv
05-29-2009, 04:18 PM
Ok, I can see where you're coming from now. I do think it's a little unfair to blame AMD for a decision Arch linux made, and the new kernel support is something most likely to be addressed in the next release (no insider info, that's just a guess). But, well, if your not satisfied by the AMD driver development cycle, I can't say much about that.
Originally Posted by Mazur
I still prefer the AMD for open source support, for regular releases (I don't upgrade to every new kernel straight away anyway, although I am running the latest driver on a 2.6.29 kernel), and for their binary driver not being quite so invasive.
Anyway, I won't bug anymore, and thankyou for letting me know!
05-29-2009, 05:57 PM
Just want to add that those drivers was removed from repository because of unsupported kernel and back then even xorg 1.6 was not supported.
Anyway I would like to ask something, are you devs going to release some more code for 3d support on open drivers?
05-29-2009, 06:12 PM
There's quite a bit of activity in the open source 3d driver trees, but most of it is in the -rewrite branches; radeon-rewrite for 1xx-5xx and r6xx-rewrite for 6xx-7xx :
The main focus for the radeon-rewrite branch is getting it ready for merging into mesa master, and that seems to be getting close. While that code stabilizes, we're trying to get the 6xx-rewrite branch ready for merging as well, so hopefully soon we can get all of the ATI-related work back into a single branch.
In case you're wondering why the radeon-rewrite branch is so important, radeon-rewrite is a single code base which can work with either the new KMS/GEM/TTM/DRI2 drm functions or with the older drm functions. This makes it a critical part of the transition to the new driver architecture.
Since most of the "not-yet-implemented" 3d functionality requires GEM/TTM as a pre-requisite, and since KMS and GEM/TTM are tied together in the current implementation, getting radeon-rewrite stabilized and into master is a pre-requisite for most of the other 3d driver improvements.
Last edited by bridgman; 05-29-2009 at 06:33 PM.
05-29-2009, 06:36 PM
I have made the decision to jump the shark and move to nvidia 9800GT for decoding and displaying hi-def content. I hung out for more than a year waiting for this functionality to come through on the ATI platform. Bought 2 motherboards 690G and 780G in preparation. Used the open source drivers, saw them improve, but couldn't wait any longer to use it for what I need, a HTPC. See ya!
05-30-2009, 08:28 AM
KMS and 3d are most important things for me for now. From that what I see nvidia is not going to support KMS in the future so AMD, it is your chance to improve .
Last edited by Mazur; 05-30-2009 at 09:10 AM.
06-05-2009, 12:10 PM
I have noticed that under 9.3 and 9.5 drivers, in windowed OpenGL apps running under Compiz, it is not possible to use Anti-Aliasing.
This seems somewhat reasonable, if the GPU is managed exclusively by the compoziting manager, but I'd like to have this confirmed by someone 'in the know'.
So my question - provided this is possible on the HW level, is it possible with the current software stack? Are some specific xorg/dri/randr/etc. versions needed? Special Xorg or /etc/amd settings?