Announcement

Collapse
No announcement yet.

2D performance on 'radeon' vs 'radeonhd'

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

  • surfer
    started a topic 2D performance on 'radeon' vs 'radeonhd'

    2D performance on 'radeon' vs 'radeonhd'

    Hi,
    I built the current repository versions of 'radeon' and 'radeonhd' and compared the 2D performance on my X1800.

    As you can see, 'radeonhd' blows 'radeon' away on the GtkDrawingArea-Text test, where 'radeonhd' is about 3X faster. I believe there is a serious problem with this aspect of the 'radeon' driver. Conversely, for the GtkDrawingArea-Circles test, 'radeon' is about 3X faster.

    A couple months ago I remember running the same type of comparison and 'radeon' performed better on the same text drawing test, but I don't know when/how the reduced performance started.

    Does anybody have any thoughts or explanation on this?

    EDIT: These were the only options I used...
    Code:
    Option "AccelMethod" "EXA"
    Option "MigrationHeuristic" "greedy"
    Yes, current repository version of DRM.

    Code:
    X.Org X Server 1.5.99.1
    Release Date: 5 September 2007
    X Protocol Version 11, Revision 0
    Build Operating System: Linux 2.6.25-zen0 i686 
    Current Operating System: Linux shocked 2.6.25-zen0 #1 SMP PREEMPT Thu May 1 20:29:16 PDT 2008 i686
    Build Date: 01 May 2008  09:21:00PM
    Radeon:



    RadeonHD:

    Last edited by surfer; 05-02-2008, 05:08 AM.

  • gamel
    replied
    2D performance on 'radeon' vs 'radeonhd'

    Next up on our testing schedule let's spend some time taking a look at power consumption. Rather than isolate the power consumption of the individual component being reviewed, we'll instead be taking a look at the power consumption of our test system as a whole to get a more real-world example of how an average system will behave in this discipline, comparing power usage with both a single Radeon HD 4850 board as well as with two boards running in CrossFire in place.For this review, we've split our power consumption testing into three segments, measuring the power used first when idling at a Windows Vista desktop, followed by testing again while playing back a Blu-Ray movie and then running an intensive 3D rendering task courtesy of 3DMark Vantage's two GPU tests.

    Leave a comment:


  • oibaf
    replied
    Originally posted by sastraxi View Post
    Now if only someone would create bleeding-edge git .deb files for those packages...

    (hopeful)
    Look here:
    http://www.phoronix.com/forums/showthread.php?p=32602

    Leave a comment:


  • sastraxi
    replied
    Now if only someone would create bleeding-edge git .deb files for those packages...

    (hopeful)

    Leave a comment:


  • bridgman
    replied
    Ahh, now that's more like it

    Leave a comment:


  • oibaf
    replied
    Update

    I updated drm modules, libdrm, mesa to latest git and tried ati driver with EXA.

    ati driver git 20080515 with EXA
    Code:
    $ gtkperf -a -c 500
    GtkPerf 0.40 - Starting testing: Tue May 20 15:50:08 2008
    
    GtkEntry - time:  0,09
    GtkComboBox - time:  7,84
    GtkComboBoxEntry - time:  5,97
    GtkSpinButton - time:  1,06
    GtkProgressBar - time:  0,60
    GtkToggleButton - time:  3,30
    GtkCheckButton - time:  3,25
    GtkRadioButton - time:  3,42
    GtkTextView - Add text - time: 12,55
    GtkTextView - Scroll - time:  2,58
    GtkDrawingArea - Lines - time:  2,48
    GtkDrawingArea - Circles - time:  1,83
    GtkDrawingArea - Text - time:  3,00
    GtkDrawingArea - Pixbufs - time:  0,27
     --- 
    Total time: 48,24
    It's a lot faster now, comparable to radeonhd with ShadowFB. "GtkTextView - Scroll" is about 3x faster than radeonhd.

    With glxgears I get this:
    Code:
    $ glxgears 
    Warning, RV530 detected, 3D HAHAHAHAHA!!.
    Mesa program:
    -------------
    # Fragment Program/Shader
      0: MOV OUTPUT[0], INPUT[1];
      1: END
    9787 frames in 5.0 seconds = 1957.329 FPS
    9890 frames in 5.0 seconds = 1977.899 FPS

    Leave a comment:


  • TechMage89
    replied
    Are you sure DRI isn't loaded for radeon? It ususally loads to support texturedvideo with CP. The DRI stuff may be the reason for the slowdown though, because it's there, but there's no acceleration.

    Leave a comment:


  • sundown
    replied
    radeonhd runs glxgears faster than radeon here. 800+ vs 300+ FPS. Of course DRI is not loaded when under radeon, not that it can detect my R500, actually.

    I run a simple xorg.conf on radeonhd with only XAA turned on. When I replace "radeonhd" with "ati" and that's the result I get.

    How about that for a comparison?
    Or is it because it can't detect my chip?
    Last edited by sundown; 05-16-2008, 12:22 PM.

    Leave a comment:


  • oibaf
    replied
    I forgot to say I had also tried EXA acceleration (either with radeon and radeonhd) and it's really slow (I interrupted the test because it was too slow). Note that I only upgraded the display drivers; xserver, mesa, drm are all the defaults found in Ubuntu 8.04.

    Also note I performed the tests with:
    Code:
    gtkperf -c 500 -a

    Leave a comment:


  • bridgman
    replied
    Yep. Radeonhd doesn't have the latest EXA code yet, but it will soon. Our testing has shown that you also need a very recent X server (to pick up some glyph cacheing improvements) in order to get the really good performance -- does that match what you are seeing ?

    Leave a comment:

Working...
X