[MPlayer-users] VDPAU performance with an Asus AT3N7A-I

Uoti Urpala uoti.urpala at pp1.inet.fi
Tue Dec 22 13:15:09 CET 2009


On Tue, 2009-12-22 at 07:49 +0100, Reimar Döffinger wrote:
> On Tue, Dec 22, 2009 at 12:53:11AM +0200, Uoti Urpala wrote:
> > On Mon, 2009-12-21 at 23:02 +0100, Reimar Döffinger wrote:
> > > With VDPAU -benchmark numbers are rather misleading.
> > > Try normal playback and give these numbers from the status line:
> > > 
> > > On Mon, Dec 21, 2009 at 10:52:44PM +0100, Richard Hartmann wrote:
> > > > V: 208.5   0/  0 14% 33%  0.0% 0 0
> > >                    ^^^^^^^
> > > (the first two % numbers).
> > 
> > I think the overall benchmark runtime (decoding 208 seconds of video in
> > 103 realtime seconds) tells more reliable information than the status
> > line numbers. The status line numbers don't really tell anything more
> > about overall time usage.
> 
> -benchmark disables sleeping and thus results in completely bogus CPU usage
> for VOs that wait for vsync.

I'm aware of that issue of course. However your status line suggestion
is not appropriate for hardware decoding, and would have been useless
even with software decoding given that the benchmark _already_ showed 2x
realtime decoding speed.

For the reasons given in my previous post the percentages reported for
hardware decoding can be _either_ too low _or_ too high. You can't say
anything definite about them. If your display FPS is significantly
higher than the video FPS then the overall benchmark time gives a
reasonably robust answer to the question "can my system play this
noticeably faster than realtime"? It doesn't give a reliable split
between "VC" and "VO" parts of time, but current status line calculation
method doesn't give that either, and it's not necessarily even a
meaningful question with hardware decoding.

His benchmark already showed an overall decoding speed of about 2x
realtime. What good would your suggestion have done even in the software
decoding case where you can get more accurate values with standard-speed
playback? The vsync issue can only make things appear slower, not
faster. If it had changed from 2x to 3x what difference would that have
made?



More information about the MPlayer-users mailing list