On Sun, 2005-12-04 at 14:48 +0100, Attila Kinali wrote:
> On Mon, 28 Nov 2005 10:51:29 +0100
> Daniel Rozsnyó <[EMAIL PROTECTED]> wrote:
> 
> > If the FPGA is enough powerfull to decode 1080i MPEG2 (is it?), then the 
> > speed of PCI is not a bottleneck.. (or at least not for me)
> 
> If you mean iDCT, yes that would be possible, if we add a
> few commands that help to perform iDCT. But be aware that
> MPEG (or any other of the current video codecs) are not
> finished after the iDCT, but you also have to perform a
> motion compensation. As far as i know, noone of the cards
> that support iDCT in hardware support also motion compensation
> in hardware. 

I think radeons do, but then, it's never been fully documented.

> Thus, the host cpu has to perform the motion compensation
> somehow, either in video memory oder in host memory.
> (yes, that's trading cpu cycles with io bandwidth)
> 
> I have never looked at this, but i do not think that we should
> waste the space we have on the FPGA for either iDCT or
> motion compensation. Most CPUs these days are fast enough
> to perform iDCT fast enough that there is no advantage
> in doing it on a card. And doing motion compensation
> in hardware would be IMHO quite expensive (logic wise).
> 
>                       Attila Kinali
> 

_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to