[MPlayer-dev-eng] DMA or not DMA

Amaury Jacquot sxpert at esitcom.org
Wed Nov 24 14:35:08 CET 2004


Arpi wrote:

> we done some initial tests last weekend, and found that a 3ghz
> p4 is fully loaded by playing 4 streams over mga_vid.
> (agp mga_vid was 17% cpu usg, pci ones eat 25% each)
> using fbdev in 32bpp, only 2 playbacks possible, at 16bpp only 3.
> the decoding was only 9% cpu per stream, the rest was eaten up by
> video data transfers. so dma could help here a lot.
> 
> (unfortunatelly hw mpeg2/mpeg4 decoder cards arent a good alternative,
> as we need to syncronize the outputs very well.)

maybe you're using the wrong kind of machine...
you should try a bi-opteron box instead of that lowly P4 machine. it 
should be fast enough for just about anything...
the problem here is not the P4 but that the processor to northbridge 
bandwith is maxed out. in effect, the same link is used to connect the 
processor to the memory and the PCI/AGP busses.

remember, the pci bus on a machine can only do 1056Mbit/s, and a 
noncompressed 8 bit 4.2.2 video is about 165Mbit/s

if you use a dual opteron box, you'll have tons more bandwith available 
to the pci/pci-X busses and it should work without a problem.

> A'rpi / MPlayer, Astral & ESP-team




More information about the MPlayer-dev-eng mailing list