[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: Fedora ATi driver performs much worse than the vanilla code

On 04/03/09 23:50, Joshua C. wrote:
2009/3/4 Roberto Ragusa<mail robertoragusa it>:
Joshua C. wrote:

I did some test and here are the results:

On the subject of performance...
  I'm running Fedora rawhide right up to date with

Under regular desktop usage I'm noticing
  * With compiz running, various special effects like wobbly windows
have a relatively low CPU impact -- good acceleration. This has been fixed in the last few weeks (thanks Dave)
  * glxgears is ok (getting 1800 ish on a V5200/X1400) - could be a bit
better but just fine.
  * Some seemingly minor activity like firefox's rotating circle that
appears in a tab whilst loading a page hit the CPU hard (up to 40% + on
  * Another app "Personal Brain" which is relatively visual has a
similar huge impact
* The latest flash player takes lots of cpu, getting up to 50% (ie 1 cpu maxed out) at < 1/4 size (default BBC iPlayer desktop size) at fullscreen it struggles -- with jerky video
  * mplayer/wmv appears to work significacantly better

So it's a real mix of working & not so working ... and not clear whether the app or driver (or mesa, X etc) is to blame...

The system is a core duo " 2.16Ghz, 3Gb ram, V5200 FireGL ie R500 series
with 256Mb, 3200x1600 virtual (large desktop across 2 screens)
EXA acceleration is enabled
 From log I note - page flipping disabled, dma 4 xv enabled, color
tiling enabled

I have a few questions

* Apart from glxgears are there any good "benchmark" apps that can be
used to help check for any regressions between updates? glxgears gives a very basic working/not working, but something that's more typical of how firefox etc drives X

* Is there any good way of concisely capturing the X/ati workload
pattern that may assist in identifying specific impacts of that workload type as it might relate to various config parameters

* Are there some obvious settings I may have missed.

* Is anyone else seeing this...

Given the above I might have enough to raise a bug... but currently am
not sure if it's working as designed. Even if not I realise the info is
far too vague.

bugs cherrybyte me uk

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]