With DXVA in MPC-HC in Windows, I watch H.264 1080p movies with 3%-5% CPU utilization (on a Radeon 4870). Without DXVA it's 80%.
Is this what we're talking about here?
Yes, you should notice similar results on Linux.
Originally Posted by RealNC
However one should bear in mind that bitstream decoding is generally fairly unforgiving of differences in video encoding. For bitstream encoding many x264 optimizations that people apply "break" bitstream decoding, so one has to bear that in mind. I suspect that for many videos it might be better just to do mc and idct and leave decode for the cpu.
Your machine is definitely powerful enough for 720p (unless it's ridiculously high bitrate and encoded with extreme h.264 options). latest mplayer with ffmpeg should work just fine. Just make sure to select your video renderer correctly, and don't run compiz. You should however get an optimized build of ffmpeg for the P4 as it helps in all the filters you would use like sharpening, deinterlacing, etc. (It doesn't help actual decoding though because of the assembly used from what I understand). ffmpeg works just fine on 720p movies on my little P-M 1.6Ghz. On the other hand, for 1080p I have to resort to CoreAVC on all the machines I have.
Originally Posted by n0nsense
Really? Ever tried Radeon driver on R500?
Originally Posted by Vighy
THAT Compiz and Textured Video magically works together! So where is this architecture change come from?
I think Vighy was talking about tearing, which requires vblank timing be pushed from drm back up to the X driver, or that *some* command streams from the X driver be blocked by drm until vblank without holding up the *other* rendering. The best way to do that is still being debated, and general consensus is that the right way will be solved in Compiz as much as in the driver (since the real solution is page flipping after composition then flow control back up to the video player apps so that they skip & double frames as needed to stay synced with the compositor's page flipping).
Originally Posted by FunkyRider
The architectural change you are talking about is the ability for Compiz to redirect content drawn via OpenGL and direct rendering (ie Redirected Direct Rendering, which needs DRI2, which needs TTM/GEM, which is being done together with KMS).
Looks like they won't be blowing by them quite yet though, since Nvidia has their own acceleration now, though it is close source. Obviously if both are fairly on par with each other feature/quality/performance wise, I'd choose AMD or Intel because of the open source feature.
Originally Posted by dashcloud
The article says:
"From the mplayer source-code run configure with the --enable-xvmc and --with-xvmclib=AMDXvBA arguments."
Doesn't work here. I get this:
Checking for XvMC ... yes (using XvMCW)
and the resulting build can't use XvMC at all. Is "AMDXvBA" the right value for the --with-xvmclib option?
(mplayer-1.0_rc2_p28058 with Catalyst 8.12 on Gentoo AMD64 here.)