Announcement

Collapse
No announcement yet.

AMD: Experimental "ShadowPrimary" Acceleration

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • phoronix
    started a topic AMD: Experimental "ShadowPrimary" Acceleration

    AMD: Experimental "ShadowPrimary" Acceleration

    Phoronix: AMD: Experimental "ShadowPrimary" Acceleration

    Hidden away within the new Catalyst 12.6 Linux driver are 2D acceleration improvements. These improvements aren't visible by default but there's a special command to activate this "ShadowPrimary" support...

    http://www.phoronix.com/vr.php?view=MTExMTY

  • pandev92
    replied
    It run on legacy driver 12.6 beta but I can't see the improvements ehehe

    Leave a comment:


  • ssvb
    replied
    Originally posted by ssvb View Post
    OK, the new catalyst 12.6 configured as "aticonfig --set-pcs-u32=DDX,ShadowPrimary,1" indeed seems to perform better when trying to scroll http://www.phoronix.com in firefox using a mouse wheel.
    Unfortunately I have to take this back. After trying to actually use the new catalyst 12.6 drivers for real, the 2D performance still feels bad. Benchmarking with cairo traces shows that the performance has not really changed (since 12.2). Don't know, maybe I just failed to activate this "ShadowPrimary" thing properly...

    Leave a comment:


  • ssvb
    replied
    Nasenbaer: thanks for the hint! It's a band-aid, but probably can help to survive until opensource drivers get good OpenCL support.

    Leave a comment:


  • Nasenbaer
    replied
    Originally posted by ssvb View Post
    OK, the new catalyst 12.6 configured as "aticonfig --set-pcs-u32=DDX,ShadowPrimary,1" indeed seems to perform better when trying to scroll http://www.phoronix.com in firefox using a mouse wheel. This was one of the worst nerve breaking experiences before, which even made me develop a habit dragging a scroll bar instead (during my previous short attempt using ati catalyst drivers). Maybe I should run some benchmarks to get the performance numbers
    A hint on improving the scolling experience in firefox: In the advanced tab in the firefox options dialog you can disable the "smooth scrolling" option. This doesn't look that nice but it stops the "chewing gum" experience when scrolling - hope you know what I mean.

    Leave a comment:


  • ssvb
    replied
    OK, the new catalyst 12.6 configured as "aticonfig --set-pcs-u32=DDX,ShadowPrimary,1" indeed seems to perform better when trying to scroll http://www.phoronix.com in firefox using a mouse wheel. This was one of the worst nerve breaking experiences before, which even made me develop a habit dragging a scroll bar instead (during my previous short attempt using ati catalyst drivers). Maybe I should run some benchmarks to get the performance numbers

    Leave a comment:


  • Linuxhippy
    replied
    It would be interesting to get some benchmark results with this option turned on.
    ... Michael?

    Leave a comment:


  • devius
    replied
    Originally posted by ssvb View Post
    Still fglrx managed to perform worse than CPU rendering in all the benchmarks with cairo traces. (...) If fglrx now allows disabling their good for nothing 2D hardware acceleration, then their drivers can become usable for normal linux desktop.
    That's good to hear. Too bad that users of older integrated graphics (like in 780G, 785G, 880G, etc) will never benefit from it. I had a 785G motherboard and was running on the IGP and it was terrible on the desktop. The 2D desktop performance was slightly (very very slightly) better than an Atom N270 with GMA950. The 3D performance was as expected and actually quite decent for what it was.

    Leave a comment:


  • d2kx
    replied
    With Chrome and fglrx, I get a massive speed improvement when I enable "disable GPU Vsync" in about:flags.

    Leave a comment:


  • ssvb
    replied
    My browser is Firefox. And in fact they are now allowing to disable the use of XRENDER accelerated 2D graphics in favour of client side software rendering, just because of bad drivers such as fglrx. The comments in that Mozilla blog provide the performance numbers for some kind of rotating psychedelic thing and fishtank. These are the workloads more suitable for GPU, and can outperform software rendering by a huge margin when having a good driver. In fact fishtank is the most GPU-friendly test from all the cairo traces used for cairo benchmarking: http://ickle.wordpress.com/2012/04/0...nce-on-radeon/

    Still fglrx managed to perform worse than CPU rendering in all the benchmarks with cairo traces. Also the users feedback from the Mozilla blog confirm that fglrx has been always the worst choice. It is important to remember that unlike synthetic tests, the typical 2D workloads usually have less scaling and almost never rotation. And such simple non-scaled 2D operations can be handled by software rendering very well.

    If fglrx now allows disabling their good for nothing 2D hardware acceleration, then their drivers can become usable for normal linux desktop.
    Last edited by ssvb; 06-01-2012, 05:49 PM.

    Leave a comment:

Working...
X