Results 1 to 8 of 8

Thread: xf86-video-ati support of 5xxx cards?

  1. #1
    Join Date
    Jun 2010
    Posts
    61

    Default xf86-video-ati support of 5xxx cards?

    Hello,

    I was under the impression that the ATI open source drivers did not support 5xxx cards. Today I stumbled upon the following page:
    http://wiki.archlinux.org/index.php/ATI

    It says that radeon xf86-video-ati driver supports 5xxx cards, but just 2D.

    My questions is, does that mean it supports a desktop with compiz? I have an HD5850 card using the proprietary catalyst driver and suffering from tearing. I'm wondering if the radeon card would resolve my issue. I don't do any gaming, I just want a usable tear-free desktop environment.

    Thanks

  2. #2
    Join Date
    Jun 2009
    Posts
    2,932

    Default

    First of all, Compiz does not work without 3d acceleration. It doesn't need very advanced OpenGL stuff, but it does need it to be accelerated.

    Secondly, 2d on 5xxx works, but to the best of my knowledge, it is unaccelerated. You need accelerated 2d for video scaling, colour conversion, resizing, and any common 2d operations.

    2d and 3d acceleration on 5xxx have been coming for quite a while now, but at the moment the open drivers are really slow at basically everything. X will work with a non-compositing window manager, but playing videos or resizing windows will be very slow.

  3. #3
    Join Date
    Jun 2010
    Posts
    61

    Default

    Thanks for clearing that up.

  4. #4
    Join Date
    Feb 2010
    Posts
    22

    Default

    I've had a chance to test a laptop with a 5xxx series card.
    And honestly i've been... well... surprised.

    Modesetting works and Xorg fallbacks to shadowfb which is quite efficient in 2D desktop tasks.
    Moving, switching windows & desktops are actually snappier than EXA was a few years (months?) ago.
    But no 3D accel at all, indeed. In KDE you could rely on soft render for compositing, it's not that bad.

    In fact, the most annoying missing feature is Xv. Not that X11 isn't fast enough these days, but you cannot manage tear-free playback without video accel (or vsynched compositing)

  5. #5
    Join Date
    Oct 2007
    Location
    Toronto-ish
    Posts
    7,514

    Default

    Yeah, I sometimes wonder if we should just drop 2D acceleration and try to get a combination of shadowfb 2D, accelerated Xv and accelerated 3D working. You'd need a compositor to bring together 2D data from system ram with 3D data in video ram, but it might be a good combination.

    2D acceleration hardware has been disappearing from GPUs for a while now; Evergreen will be the 3rd generation of ATI hardware that uses the 3D engine for 2D accel.

  6. #6
    Join Date
    Feb 2010
    Posts
    22

    Default

    What struck me in the early days of EXA were lacks in desktop responsiveness. With XAA and shadowfb, it's always been fast and sharp with much less CPU load.
    This said, being just an end user, i'm probably missing parts of the big picture. And don't get me wrong, I'm quite happy now with OSS ATI.

    Btw, modesetting done through KMS, unified 3D accel (gallium)... does it sound like the end of the DDX ?...

  7. #7
    Join Date
    Dec 2007
    Posts
    2,395

    Default

    For EXA, you really need to accelerate the Composite hook since that's where most of the modern toolkit stuff ends up. In the early days of EXA, few drivers supported the Composite hook so performance sucked because you ended up ping-ponging between hw and sw rendering which almost always performs worse than pure sw or pure hw.

    At this point XAA is mostly software rendering as it's so bit rotten at this point, all it accelerates is on-screen blits and upload to screen. As such, most stuff ends up purely in software, so it performs more like shadow fb.

  8. #8
    Join Date
    Oct 2007
    Location
    Toronto-ish
    Posts
    7,514

    Default

    Quote Originally Posted by dmfr View Post
    Btw, modesetting done through KMS, unified 3D accel (gallium)... does it sound like the end of the DDX ?...
    That was the initial thinking, but right now there are enough differences between the memory management hardware on different GPUs that the memory management APIs are still significantly different between drivers. They're nearly all using GEM, but GEM allows driver-specific interfaces.

    As long as that is the case (and AFAIK there are no plans to change it) we're more likely to see GPU-specific DDXes continue. For the short-to-medium term, where ongoing support of UMS is still necessary for enterprise distros, the GPU-specific DDXes are pretty much a necessity.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •