GeForce 8200, VDPAU: 1080p is a no-go

[D-G]

I use Ubuntu 10.04 with the "official" 195.36.15 drivers and corresponding libvdpau as offered by the OS. The GPU is a GeForce 8200 on an ASUS M3N78-VM motherboard. Compiz is deactivated. Composite Extensions are deactivated.

MPlayer was self-compiled with the mplayer-build scripts. FFmpeg-mt is used.

Well, to cut a long story short, when using VDPAU, I can play 720p perfectly fine, but I can't play any 1080p material flawlessly. The audio comes out alright, but the video lags behind severely, displaying at something like 4-5 fps, with MPlayer displaying the dreaded "Your computer is too slow to play this" message appearing in the terminal window. The videos are all well within sane specifications. High profile, level 4.1, at most 4 reference frames. But even simple 1080p movie trailers from Apple don't work.

Software decoding works.

Can't really believe that this is an MPlayer issue, since the 1080p videos play fine on Windows 7 when using DXVA in MPC-HC.

[cehoyos]

Originally Posted by D-G

MPlayer was self-compiled with the mplayer-build scripts. FFmpeg-mt is used.

Since Apple's 1080 trailer should be no problem, could you test:

$ svn co svn://ffmpeg.org/mplayer/trunk MPlayer
$ cd MPlayer
$ ./configure && make
$ ./mplayer atrailer.mov

If that doesn't work, please post complete, uncut output of the last command. (Note that some trailers show constant A/V desync, but that is unrelated to VDPAU.)

And please post qvdpautest output.

Carl Eugen

[Stephen Warren]

If you have an AMD CPU, please be aware that when the CPU is running at a low speed (due to the decoding load being shifted to the GPU) the bus frequency will also be lowered, which reduces the memory bandwidth available to the GPU (since an 8200 is a shared memory system), which negatively affects decoding performance.

If you lock your CPU frequency to a higher value, this may alleviate the symptom.

Search this forum or Google for more details.

[hl_]

By the way, this is only a problem on older Athlon64 (K8) CPUs. Newer Phenom/AthlonII (K10) CPUs run the IMC at full speed independent of CPU power state.

[gradinaruvasile]

Well i have a 8200 on ASUS M3N78-VM and 1080p is just fine. I had a 3200+ Athlon @2GHz that scaled back to 1 GHz and 1080p was running perfectly well, no hiccups (~10% CPU).
Now i got a 250 dual core that runs on 800 MHz on light load and 1080p is the same if not a bit faster/snappier (this is subjective) at ~5% CPU .

I use Debian Squeeze. The thing is that Debian has the vdpau-enabled mplayer in its repos from the get-go, no ppas and compiling required.

[D-G]

Thanks for all the replies.

Originally Posted by cehoyos

Since Apple's 1080 trailer should be no problem, could you test:
(snipped)
If that doesn't work, please post complete, uncut output of the last command. (Note that some trailers show constant A/V desync, but that is unrelated to VDPAU.)

Have done. Played a lot better than with the FFmpeg-mt build, but it still stuttered. Output below:

MPlayer SVN-r31955-4.4.3 (C) 2000-2010 MPlayer Team
mplayer: could not connect to socket
mplayer: No such file or directory
Failed to open LIRC support. You will not be able to use your remote control.

Playing /home/thr/Downloads/111401738.mov.
libavformat file format detected.
[lavf] stream 0: audio (aac), -aid 0, -alang eng
[lavf] stream 1: video (h264), -vid 0
VIDEO:  [H264]  1920x816  24bpp  23.976 fps  9827.1 kbps (1199.6 kbyte/s)
Clip info:
 major_brand: qt  
 minor_version: 537199360
 compatible_brands: qt  
 title: "TRON: Legacy" Trailer #2
 title-eng: "TRON: Legacy" Trailer #2
 artist: Yahoo! Movies
 artist-eng: Yahoo! Movies
 copyright: © 2010, Walt Disney Pictures.
 copyright-eng: © 2010, Walt Disney Pictures.
==========================================================================
Forced video codec: ffh264vdpau
Opening video decoder: [ffmpeg] FFmpeg's libavcodec codec family
Selected video codec: [ffh264vdpau] vfm: ffmpeg (FFmpeg H.264 (VDPAU))
==========================================================================
==========================================================================
Opening audio decoder: [faad] AAC (MPEG2/4 Advanced Audio Coding)
AUDIO: 48000 Hz, 2 ch, s16le, 187.2 kbit/12.19% (ratio: 23404->192000)
Selected audio codec: [faad] afm: faad (FAAD AAC (MPEG-2/MPEG-4 Audio))
==========================================================================
AO: [pulse] 48000Hz 2ch s16le (2 bytes per sample)
[Mixer] No hardware mixing, inserting volume filter.
Starting playback...
[VD_FFMPEG] Trying pixfmt=0.
Movie-Aspect is undefined - no prescaling applied.
VO: [vdpau] 1920x816 => 1920x816 H.264 VDPAU acceleration 
[VD_FFMPEG] XVMC-accelerated MPEG-2.
A:   4.5 V:   2.8 A-V:  1.684 ct:  0.022   0/  0 95% 47%  2.1% 50 0 


           ************************************************
           **** Your system is too SLOW to play this!  ****
           ************************************************

And please post qvdpautest output.

Here it is, with the CPU at 1 GHz ...

qvdpautest 0.5.1
AMD Athlon(tm) Dual Core Processor 4850e
NVIDIA GPU GeForce 8200 (C77) at PCI:2:0:0 (GPU-0)

VDPAU API version : 1
VDPAU implementation : NVIDIA VDPAU Driver Shared Library  195.36.24  Thu Apr 22 19:52:55 PDT 2010

SURFACE GET BITS: 648.373 M/s
SURFACE PUT BITS: 819.492 M/s

MPEG DECODING (1920x1080): 38 frames/s
MPEG DECODING (1280x720): 100 frames/s
H264 DECODING (1920x1080): 35 frames/s
H264 DECODING (1280x720): 60 frames/s
VC1 DECODING (1440x1080): 42 frames/s
Profile unsupported.

MIXER WEAVE (1920x1080): 205 frames/s
MIXER BOB (1920x1080): 315 fields/s
MIXER TEMPORAL (1920x1080): 68 fields/s
MIXER TEMPORAL + IVTC (1920x1080): 29 fields/s
MIXER TEMPORAL + SKIP_CHROMA (1920x1080): 96 fields/s
MIXER TEMPORAL_SPATIAL (1920x1080): 22 fields/s
MIXER TEMPORAL_SPATIAL + IVTC (1920x1080): 13 fields/s
MIXER TEMPORAL_SPATIAL + SKIP_CHROMA (1920x1080): 25 fields/s
MIXER TEMPORAL_SPATIAL (720x576 video to 1920x1080 display): 86 fields/s

MULTITHREADED MPEG DECODING (1920x1080): 46 frames/s
MULTITHREADED MIXER TEMPORAL (1920x1080): 36 fields/s

,,, and with the CPU at 2.5 GHz:

qvdpautest 0.5.1
AMD Athlon(tm) Dual Core Processor 4850e
NVIDIA GPU GeForce 8200 (C77) at PCI:2:0:0 (GPU-0)

VDPAU API version : 1
VDPAU implementation : NVIDIA VDPAU Driver Shared Library  195.36.24  Thu Apr 22 19:52:55 PDT 2010

SURFACE GET BITS: 630.852 M/s
SURFACE PUT BITS: 803.511 M/s

MPEG DECODING (1920x1080): 63 frames/s
MPEG DECODING (1280x720): 137 frames/s
H264 DECODING (1920x1080): 55 frames/s
H264 DECODING (1280x720): 104 frames/s
VC1 DECODING (1440x1080): 73 frames/s
Profile unsupported.

MIXER WEAVE (1920x1080): 206 frames/s
MIXER BOB (1920x1080): 324 fields/s
MIXER TEMPORAL (1920x1080): 68 fields/s
MIXER TEMPORAL + IVTC (1920x1080): 42 fields/s
MIXER TEMPORAL + SKIP_CHROMA (1920x1080): 96 fields/s
MIXER TEMPORAL_SPATIAL (1920x1080): 22 fields/s
MIXER TEMPORAL_SPATIAL + IVTC (1920x1080): 16 fields/s
MIXER TEMPORAL_SPATIAL + SKIP_CHROMA (1920x1080): 25 fields/s
MIXER TEMPORAL_SPATIAL (720x576 video to 1920x1080 display): 85 fields/s

MULTITHREADED MPEG DECODING (1920x1080): 48 frames/s
MULTITHREADED MIXER TEMPORAL (1920x1080): 36 fields/s

Originally Posted by Stephen Warren

If you have an AMD CPU, please be aware that when the CPU is running at a low speed (due to the decoding load being shifted to the GPU) the bus frequency will also be lowered, which reduces the memory bandwidth available to the GPU (since an 8200 is a shared memory system), which negatively affects decoding performance.

If you lock your CPU frequency to a higher value, this may alleviate the symptom.

It did indeed. Now comes the dumb question: why? I just booted into Windows 7, opened CPU-Z and watched the CPU frequency while playing back the same 1080p trailer from Apple I used above. If CPU-Z is to be believed, the frequency remained at 1 GHz at any given time, yet the trailer played flawlessly.

Also, according to qvdpautest, as presented above, my card manages 35 frames/s when decoding H.264 while the CPU frequency is 1 GHz. Given that the trailer is 23.976 frames/s, my overly naive math tells me that I shouldn't be presented with stuttering ...?

Ah, I forgot, I'm using the 64-bit version of Ubuntu. Don't know if that would make any difference.

[Stephen Warren]

Could you run nvidia-settings and look at the PowerMizer page while decoding the video, and running the CPU at 1GHz and 2.5GHz. Is the selected PowerMizer level different between the two runs?

It's odd that you get-/put-bits and mixer performance are essentially identical at both CPU frequencies, but the decoder performance varies.

[gradinaruvasile]

I just played the tron legacy trailer in 1080p from

http://www.youtube.com/watch?v=-xxLw7S6EaA

(from youtube) with smplayer - i let it to cache all the file, go into /tmp folder and copy the file from there (has a random name, but you see the icon), it can be played with any movie player.

M3N78-VM mobo (using the onboard 8200), Athlon II 250 @3.0GHz processor, 2 GB DDR2 @800 MHz RAM, Debian Squeeze 32-bit, smplayer/mplayer from the debian multimedia repo, nvidia drivers 256.44 from the nvidia site.
The playback is perfectly smooth, no hiccups or anything, mplayer eats 3-5% CPU, smplayer 3-5%, the CPUs stay at 800 MHz.

Id say the problem problem related to your ffmpeg/mplayer.

Edit:

I have adaptive clocking anabled, but only 1 performance level exists and no options to change anything related to it. I have a laptop with 135M card and that has 3 levels and i have the option to choose adaptive clocking or maximum performance. But this card dont seem to have performance levels.
Also the memory clock is shown as 400, i remember sometimes back (dont know exactly what driver version) seeing 800 at its speed.

[D-G]

Originally Posted by Stephen Warren

Could you run nvidia-settings and look at the PowerMizer page while decoding the video, and running the CPU at 1GHz and 2.5GHz. Is the selected PowerMizer level different between the two runs?

Just a single PowerMizer level is displayed in nvidia-settings, I can't change it. And it also remains the same all the time, whether the CPU runs at 1 GHz or 2.5 Ghz.