gl.videoplane texture size
I’m building a patch that renders 6 videoplanes to a single window.
I’m hopinng to work with short animation-compressed qt’s with alphas stored in RAM for speed. Hopefully at around 512×512 size.
I’m hoping for a frame rate of 30fps. I’m getting about 25fps.
I’m running a dual g5 2.0ghz with an ATI 9800 card.The CPU activity monitor is running at about 50% per processor.
I’ve been testing with textures that are 256 x 256, mostly tiff files with alpha channels.
Reducing the size of the jit.qt.movie dimensions seems to increase the frame rate a little, whereas raising drops the frame rate.
Where are the big bottlenecks?
If you can run 4 dv movies at full speed, surely 6 movies is no biggie.
Will look at colourspace options, but are there any other speed bumps?
Should I be getting more than this? Am I overloading the gl card? Surely not.
I am running 6 x fft’s on audio at the same time, but this seems to be negligable.
Your bottleneck is almost certainly how much data you can reasonably expect to get off the hard drive. I think playing 6 simultaneous DV files is difficult even for a RAID system.
Try opening 6 DV files in QuickTime and see how your hard disk holds up.
Having said that, DV codec is also CPU-intensive. If you really need 6 files playing simultaneously, try compressing the files with a different codec (for example, Photo JPEG looks acceptable at low data rates and is fast to decompress).
In addition to what Kurt suggested, you should try using loadram for
your movies if they are short enough. This will solve any disk reading
As mentioned several times, using uyvy colorspace will vastly improve
performance, as well as doing all pixel processing using jit.gl.slab.
allthough recently on a quad g5 with a 10.000 rpm extra hd i had no
harddisk problems playing back 6 dv files at the same time through
jitter, to 3 independant vga outputs ..
I’m getting that framerate while just using the 256 x 256 tiff files, not qt movies.
And been doing the loadram thing when using movies. So it ain’t the the drive read speed.
There’s something else going on here.
Will do the uyvy thing. If it’s still slow, I’ll post the patch.
Some other thoughts.
1)I’m sending a separate trigger to each videoplane but ‘automatic’ is still set to the default on. Couldn’t see how this setting worked. Am I doing it wrong?
2) Ideally my texture size would be 720 x 576. I went with that whole power of 2 thing but I hear that it doesn’t really matter about that. Is this texture size realistic with 6 movies, if playing back from RAM?
3) Although the CPU monitor is reading about 50%, I get the feeling that it’s this causing the slowdown rather than the GPU. Currently I’m only doing sound analysis and reading the quicktimes on the CPU. No colour stuff or anything. But the CPU usage goes up considerably by increasing the texture size. Is there anything (apart from uyvy) that optimises that ‘getting the qt movie onto the videoplane’ process?
4) Have I just got the world’s slowest G5? Or sloppy Max skills?
(don’t answer that one).
I have simular feeling on projects I worked out on PCs too. Usually the CPU
usage raise up to 50% (for one of the logical unit on my intel Pentium) and
I can see the performance starts to drop.
I remeber right, seems Max only utilize two threads, high-priority and
low-priority, this may be the reason…