[MPlayer-users] Playing large frame-size uncompressed sequences

Dean S. Messing deanm at sharplabs.com
Tue Jan 28 22:21:12 CET 2003


Thanks, Richard, for the help.
Comments interspersed.

 :: On Mon, Jan 27, 2003 at 09:12:38PM -0800, Dean S. Messing wrote:
 :: > [Automatic answer: RTFM (read DOCS, FAQ), also read DOCS/bugreports.html]
 :: > 
 :: > Gurus, 
 :: > 
 :: > Very soon our laboratory will be acquiring several sequences
 :: > of progressive HDTV data. 10 second long sequences.
 :: > 
 :: > Each frame is 1920x1440.  The data will either be
 :: > in YUV 4:2:2,  YUV 4:2:0, or RGB format, progressive, 60 fps.
 :: > 
 :: > I can convert the individual frames to any uncompressed format that
 :: > Mencoder/Mplayer would like in order for Mencoder to generate me a
 :: > playable file.
 :: 
 :: If they're RGB, probably go for PNG as input. I'm not sure what raw
 :: YUV formats mplayer supports right now, but there should be something
 :: that works.

Ok. Png it is.  Except that I've yet to get Mencoder to
encode a .png sequence to .avi so I can play it.
I presume that's what you are suggesting.  If not, please clarify.
(I'm still an rank beginner w.r.t. Mplayer/Mencoder.)

 :: > My question is: will Mplayer be able to play this in "real time"
 :: > (i.e. 60 fps) out of memory?  I'm prepared to put the 4 or 5 GBytes of
 :: > Ram in the machine and get whatever video card is necessary.  Will a
 :: > dual 1.3 GHz Athlon (or similar) machine handle the needed data rates?
 :: 
 :: Dual won't help you one bit. Go for the maximum clock you can get
 :: instead.

Yes, I realised after I sent it that dual probably would not matter
(except for keeping other parts of the system snappy during playback).
I think there was a discussion here a short time ago about this.
Ok, I'll buy the fastest processor/bus combo I can get.

I presume that Mplayer will load the 4 or 5 Gig sequence into memeory
and play it from there?  Or can it keep up with the framerate off of
disk.

Lets see (naive calc follow):

For a 1920x1440 YUV-4:2:2 frame we have

1920*1440 * 2 = 5,529,600 bytes

(The two comes from 1Y + .5U + .5V for 4:2:2 data)

At 60 fps this works out to     ~332,000,000 bytes / sec

or about 316 MBytes/secs  being pulled from memory!

Is this realistic on a fast PC?  Certainly it is not realistic
from disk.  Even the fastest SCSIs which operate in burst mode at these
rates won't sustain them off the platter, I don't think.
Correct me if I'm wrong.

 :: For max video performance and quality you should probably get
 :: a Matrox G400. That way you can use mga_vid, which will avoid some
 :: unnecessary copying from going through the X server, and updates will
 :: be synced to vertical retrace so you won't see shearing. Also the
 :: filters are said to be much higher quality than on the "gamer" cards,
 :: but I don't know much about that.

Yes, Arpi had the same suggestion couple weeks ago when I asked
about "exact rendering".  I have a G450 laying around
but it does not have DVI out which is required.  I'll have
to check on current Matrox offering with DVI-I (digital) out.

Any other thoughts (from anyone) will be appreciated.

                          Dean S. Messing
                          Display Algorithms & Visual Optimization Lab
                          Information Systems Technologies Dept.
                          Sharp Laboratories of America
                   E-Mail: deanm at sharplabs.com




More information about the MPlayer-users mailing list