[PD] lowering gem camera capture CPU overhead

Anders Friberg andersf at speech.kth.se
Sun Mar 18 20:48:00 CET 2007

Hi Ico,
I've run into exactly the same problem. Most probably it is the decoding 
of the firewire signal that eats the cpu cycles. The dv format (same as 
firewire) is compressed into a 29 Mbits/sec stream (see e.g. 
http://www.adamwilt.com/DV-FAQ-tech.html#DVformats) that needs to be 
unpacked before processing in GEM (or any other video software). My 
solution is to use analog video cameras with a capture card.  In this 
way I am able to use four simultaneous cameras using less than half of 
the cpu power on a standard winpc with pentium 4, 3GHz. I don't need 
much picture quality so I got really cheap surveillance CCD cameras 
costing less than 50 euro each connected to a four-input capture card 
(about 190 euro). Another great thing is that your are not limited in 
cable length as with firewire. Another solution would be to find a 
firewire capture card with built-in decoder (I guess there are such 
cards out there).

Ivica Ico Bukvic wrote:
> Hi all,
> I've been trying to shave off precious cycles off the FW camera input 
> capture. The problem is that vanilla capture of a FW camera feed 
> (NTSC) already introduces 50+% of CPU overhead on an AMD64 3000+. I 
> have experimented with lowered gem frame value, but anything below 20 
> seems to jagged.
> So, here are the questions:
> What is the cause of such a high capture overhead?
> Is there a way to minimize such overhead beyond just optimizing code 
> for sse and other similar methods, plus the aforesaid frame option for 
> the gem object?
> Many thanks!
> Best wishes,
> Ico
> ------------------------------------------------------------------------
> _______________________________________________
> PD-list at iem.at mailing list
> UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.puredata.info/pipermail/pd-list/attachments/20070318/d95b6813/attachment.htm>

More information about the Pd-list mailing list