[Theora-dev] Theora camera/fast computers

Ivan Popov pin at medic.chalmers.se
Wed Mar 23 07:53:21 PST 2005

Hi Andrey,

On Tue, Mar 22, 2005 at 02:35:47PM -0700, Andrey Filippov wrote:
> and this one is the full 30fps (1280x1024):
> http://www.elphel.com/3fhlo/samples/333_samples/m021_300_1_70.ogg

> will be able to  decode and render the video with the currently available
> software? Any practical recommendations? Does anybody have such a system
> that can play the
> http://www.elphel.com/3fhlo/samples/333_samples/m021_300_1_70.ogg with
> full resolution and full speed without skipping?

Testing on the best hardware I have nearby:
a dual 3 GHz P4 with nVidia GeForce 6600 and the proprietary X11 drivers.

The following sequence applies literally
to any Linux distribution with a 2.6 kernel, or FreeBSD 5.3 or NetBSD 2.0,
given an accelerated X11 driver and free 500MB on the root partition.

So, you can test available hardware yourself with the following:

[as root:]

wget http://www.konvalo.org/pub/coda-client-setup
sh coda-client-setup
 [accept defaults]

modprobe coda     ["kldload coda" on FreeBSD, no need on NetBSD]

startstopvenus start

[then as a regular user (or as root as well) under X11:]

[it takes some time depending on your net connection,
you get a new bash session]

mplayer.theora-mmx m021_300_1_70.ogg -fs
[takes some time as well, for the first run, so we do it
before the measurements]

time mplayer.theora-mmx m021_300_1_70.ogg -fs
["-fs" means fullscreen]
real    0m 10.35s        <---------------
user    0m 9.60s
sys     0m 0.02s

[which means slightly under 30bps (the original is about 10.1 sec, right?)]

[trying at 25 fps, which is the natural choice for the European market:]

time mplayer.theora-mmx m021_300_1_70.ogg -fps 25 -fs
real    0m 12.19s        <----------------
user    0m 9.16s
sys     0m 0.08s

[timing is "almost" OK]

[optionally, as root:]

sh coda-client-setup -u
[removes all traces of the coda client from the system]


 - the picture is not always smooth,
   probably because of mplayer being single-threaded -
   the peak load is not distributed over several frames,
   so one or several frames become slightly delayed when the decoder
   encounters an "extra expensive" frame
 - adding "-framedrop" to the command line does not change timing, nor
   the smoothness, i.e. the delays are under the threshhold where mplayer
   begins to decimate frames
 - this is the non-experimental decoder with the decoder-specific
   MMX patches applied. It is the fastest decoder of those which I could test.

My guess is that on a 3.3GHz hardware you will be fine for scenes like that one
but possibly not when there is plenty of motion, like an exhibition hall...

You would be fine though with even a slightly smaller resolution, due to the
quadratic reduction in complexity.


More information about the Theora-dev mailing list