danscofi

I'm running a 2.8 Quad-Core Mac Pro w/ 16GB RAM and three ATI Radeon HD 2600 XT video cards.

I need to run three HD movies simultaneously on three different screens. When I use fullscreen I drop from ~25fps to about 15fps. I know this has been discussed at length on the forums but I'm still having trouble. I'm using a modified version of Andrew's Bensons loadram patch and I've found that ProRes422(LT) @ 1280x720 works best.

I've tried to limit all unnecessary elements (although I am running eyetracking and sending osc messages to and from processing) but still the videos don't play smoothly enough.

smooth-hd-playback-on-3-screens

its a bug since years, if you use multiple gpus.

use the rect message to put the window to the gpu and cut one pixel on the right. for example [rect -1280 0 -1 720]

i had the best experience with the aic codec.

but it will not play as smooth as in the quicktime player.

perhaps cycling will give the jit.qt.movie an update.

solution:
dont use fullscreen.
use the rect message to put the window to the gpu and cut one pixel on the right. for example [rect -1280 0 -1 720]

but it will not play as smooth as in the quicktime player.
perhaps cycling will give the jit.qt.movie an update.
that would be nice.

Have you tried using OpenGL to display the video as a texture on jit.gl.videoplane?

Lee: I'll try rect approach and see how that goes.

dtr: yes i've tried both vade's jit.gl.render > videoplane "GPU - synced with movie" patch and also the "video processing system 4" employing loadram.

I'll try the rect message that lee suggested and let you guys know what happens.

dtr: yes i've tried both vade's jit.gl.render > videoplane "GPU - synced with movie" patch and also the "video processing system 4" employing loadram. 

So I tried @rect and that makes no difference in framerate--in fact it's a little slower. I'll try a matrox and see how that goes...although I have three different videos (one on each screen) and I'm pretty sure a matrox will display one giant image that is spread across three screens...

> I'm pretty sure a matrox will display one giant image that is spread across three screens...

yes, but you position your 3 video's on the triple width screen so that they exactly match each screen output (3x1280x720 for example). or you can render your 3 video's to one triple width video, if they don't need to be interactive.

You also need to ensure that your windows init OpenGL on the proper GPU, by moving them over to the proper window coordinates before turning on the metro/jit.gl.renderer so resources are allocated on the right GPU.

I was going insane with this but I figured it out. I'm now getting 60fps on three 1360x768 monitors using 1280 x 720 ProRes422(LT) with three separate nvidia cards.

Main thing I found is that dtr is right: you cannot have negative screen coordinates. You must rearrange your virtual monitor setup so that the externals are on the right and thus positive.

Special thanks to Vade, Andrew Benson and Scott Fitzgerald. Patch below.

i did some additional tests on my machine.

- i first put the windows on the output screens, then i started the jit.gl.render.

- i used a standalone application build with max 5.1.9

i did NOT get the patch working with @pos and fullscreen messages.

with the rect message i get the expected 60fps,

- if all screen coordinates are positve AND the window width is 1 pixel smaler than the output.

- if all screen coordinates are negative AND i send the messages [border 1] [border 0] before i turn jit.gl.render on AND the window width is 1 pixel smaler than the output.

if the widths of the jit.gl.window are exact the screens witdths, i get framerates of 20 fps.

no chance to get every pixel of my window to the output !!

- cc.uyvy2rgba.lite.jxs shader is involved or not. on a mac you can simply add the attributes: @automatic 1 and @colormode uyvy to the jit.gl.videoplane

- wich gpu is for the max application and which is for the output

- if you are on snow leopard or on lion (latest update)

- if you load a movie into jit.qt.movie or not

it would be great, if in future versions of max we get rid of these bugs.

for every machine we have to choose different methods to render out some frames of the gpu.

i never had this problem in openframeworks.

same hardware, same os, just playing movies on the gpu.

- mac pro 4,1
- 1 ATI Radeon HD 5770
- 3 NVIDIA GT 120
- 4 monitor attached all 1280x720
- snow leopard and lion

for every test
- i first put the windows on the output screens, then i started the jit.gl.render.
- i used a standalone application build with max 5.1.9

i did NOT get the patch working with @pos and fullscreen messages.
my framerates are 20 fps.

with the rect message i get the expected 60fps,
- if all screen coordinates are positve AND the window width is 1 pixel smaler than the output.
- if all screen coordinates are negative AND i send the messages [border 1] [border 0] before i turn jit.gl.render on AND the window width is 1 pixel smaler than the output.

if the widths of the jit.gl.window are exact the screens witdths, i get framerates of 20 fps.
no chance to get every pixel of my window to the output !!

it makes no difference if
- cc.uyvy2rgba.lite.jxs shader is involved or not. on a mac you can simply add the attributes: @automatic 1 and @colormode uyvy to the jit.gl.videoplane
- wich gpu is for the max application and which is for the output
- if you are on snow leopard or on lion (latest update)
- if you load a movie into jit.qt.movie or not

it would be great, if in future versions of max we get rid of these bugs.
for every machine we have to choose different methods to render out some frames of the gpu.

and it is solvable !!!
i never had this problem in openframeworks.
same hardware, same os, just playing movies on the gpu.

Smooth HD playback on 3 screens