ATI Driver 4 Video Playback

[bolzerrr]

Hello!

I?m looking for the best driver in order to run videos on linux.
I m running XBMC on a 4200 onboard card, actually with 256mb shared, when i run a 1080p video on 1280x(later i will use hdmi on my lcd with 1080 res) resolution i get 120%CPU, and i think this should go better.
Actual software situation:

ubuntu 9.10 with 32 bit generic-pae kernel
xorg-fglrx driver (restricted) via apt-get

no settings/configurations made

what drive would you suggest? newst fglrx?raedonhd? any speical settings/patches?

Cheers

[droidhacker]

You're looking for ATI driver version 4? How many years ago would that be?

Current AMD driver version is 10.1.

[Pfanne]

i dont know if you are trying to be ironic , but i think the 4 in the thread title stands for "for".

[RealNC]

Hello!

I?m looking for the best driver in order to run videos on linux.
I m running XBMC on a 4200 onboard card, actually with 256mb shared, when i run a 1080p video on 1280x(later i will use hdmi on my lcd with 1080 res) resolution i get 120%CPU, and i think this should go better.
Actual software situation:

ubuntu 9.10 with 32 bit generic-pae kernel
xorg-fglrx driver (restricted) via apt-get

no settings/configurations made

what drive would you suggest? newst fglrx?raedonhd? any speical settings/patches?

Cheers

The open driver doesn't support any kind of video acceleration. The only solution right now is fglrx with some extra work to get XvBA working both in the driver as well as the applications. It's a crap-fest though with lots of bugs, crashes and hangs.

[bridgman]

The open driver supports video acceleration, just not *decode* acceleration. It does accelerate the most time-consuming functions - scaling, colour space conversion etc...

[RealNC]

The most time consuming function of 1080p H.264 video (as the OP mentioned) is scaling and color space conversion? I learn something new every day, I guess

[bridgman]

Try running it with -x11 output and see what happens to your CPU usage. The exact split of CPU time depends on the CPU and memory, of course, but there are a number of analyses of playback and they tend to indicate that render accel is still the biggest time hog.

I guess that might not be true if you have a display capable of showing 1920x1080 without scaling, in which case the decode part would probably be more time-consuming than render. Maybe I have to get me one of those big displays

[RealNC]

I just tested with a 720p video. I kept the window smaller than the native video size so that there's scaling involved. Radeon OSS driver. MPlayer with ffmpeg-mt.

Xv: About 30% total CPU
X11 (slow): About... 32% total CPU

I paid attention to test the exact same part of the video both times.

Does that count or did I do something wrong? :P

[RealNC]

And I forgot to test *with* video "acceleration" (OK, hardware decoding as you mentioned it):

MS Windows DXVA: under 1% CPU
MS Windows Overlay: 35%

Unfortunately I don't know of a way to disable hardware scaling and color conversion in Windows (I'm using MPC-HC there with FFDShow.)

So I guess my point is still valid The decoding part is by *far* (orders of magnitude) more CPU consuming.

But again, I may be not testing correctly or something.

[bridgman]

What you did seems reasonable but the results seem odd.

When you say X11 (slow) do you mean that the playback was actually slow ? If so that implies the CPU couldn't keep up, presumably because a core was saturated and the scaling/CSC code was single threaded...

Just guessing though.