[vorbis] Performance tests

Greg Wooledge greg at wooledge.org
Tue Oct 16 16:33:20 PDT 2001

Erik Olofsson (erik at o3games.com) wrote:
> I will do some additional testing later today to assert this. Do you have
> any suggestions as to what decoders etc I should use?

> From: volsung at asu.edu [mailto:volsung at asu.edu] 
> As Michael pointed out, this seems quite low.  My rough tests were showing
> 30-40% CPU use w/ Pentium 166 decoding ogg.

Here's a slightly more useful measurement, performed on a K6-2 333 MHz
machine running Debian GNU/Linux (kernel 2.2.19, libc6 2.2.4).

jekyll:~$ time ogg123 -d raw -f /dev/null /mp3/Tori_Amos/Tori\ Amos\ -\ Winter.ogg 
Playing from file /mp3/Tori_Amos/Tori Amos - Winter.ogg.
real    1m11.415s
user    1m2.420s
sys     0m1.000s

That's 63.420 seconds of CPU time to decode this song.  The playing
length of the song (reported by ogginfo) is 342.066667 seconds.  Thus,
the playback "speed" is 5.39x realtime.  Or to put it another way,
realtime playback would use about 18.5% of my CPU's cycles (using the
decoder found in ogg123).

For comparison:

jekyll:~$ time mpg321 -s /mp3/Tori_Amos/Tori\ Amos-Yes,\ Anastasia.mp3 >/dev/null
High Performance MPEG 1.0/2.0/2.5 Audio Player for Layer 1, 2, and 3.
Version 0.59q (2001/Aug/27). Written and copyrights by Joe Drew.
[09:33] Decoding of Tori Amos-Yes, Anastasia.mp3 finished.

real    2m5.604s
user    1m17.310s
sys     0m0.820s

Here, CPU time is 78.130 seconds, song length is 573 seconds, so the
playback rate is 7.33x realtime, or CPU usage for realtime playback would
be about 13.6%.  (mpg321 uses Rob Leslie's MAD decoder; it's focused on
quality rather than speed.)

I'm ignoring the "real" (wall-clock) times reported, because this system
was not idle during the decoding tests.  Also the files are being retrieved
from an NFS server over a 10Mbps LAN, which is another potential source
of latency that should not be factored into the tests.

Given the relatively poor FPU performance of the K6 vs. the Pentium
family, I'm inclined to believe that volsung's numbers are accurate.

Greg Wooledge                  |   "Truth belongs to everybody."
greg at wooledge.org              |    - The Red Hot Chili Peppers
http://wooledge.org/~greg/     |

<LI>application/pgp-signature attachment: stored
-------------- next part --------------
A non-text attachment was scrubbed...
Name: part
Type: application/octet-stream
Size: 241 bytes
Desc: not available
Url : http://lists.xiph.org/pipermail/vorbis/attachments/20011016/fb019374/part-0001.obj

More information about the Vorbis mailing list