theLadLearnsMax

In a system with multiple graphics cards, does direct-to-window rendering from jit.qt.movie do the right thing when a window is moved from one display to another?

More generally, how does jitter decide which GPU to use when there are several cards?

More specifically, I am working on an installation with multiple quicktime movies on multiple displays. It's a very simple configuration with one window filling each display. Everything works fine, except there seem to be some subtle issues regarding performance vs. window placement.

While developing the patch on a laptop with a single display, there was a big performance gain when I set jit.qt.movie to render direct-to-window.

But on a MacPro with multiple video cards, it's the opposite. I am getting much better performance by banging matrices out of jit.qt.movie into jit.window objects through patchcords. I have one monitor on one card set up for a control console, and two outputs from the other card connected to two projectors. The jit.windows are aligned to the displays and do not span cards.

It also seems like toggling noaccel and/or fsaa on and off after positioning the windows affects performance. (Not intentionally changing these settings, but just toggling them). And finally, I have a hunch that setting all three displays to the same resolution makes a difference.

This all makes we wonder which GPU gets used for hardware scaling when there are multiple cards, and if there are other tricks in this kind of environment.

I am using 5.0.3. I think the fullscreen bug in earlier versions had become a feature in my patch, but that's another story!

While developing the patch on a laptop with a single display, there was a big performance gain when I set jit.qt.movie to render direct-to-window. 

But on a MacPro with multiple video cards, it's the opposite. I am getting much better performance by banging matrices out of jit.qt.movie into jit.window objects through patchcords. I have one monitor on one card set up for a control console, and two outputs from the other card connected to two projectors. The jit.windows are aligned to the displays and do not span cards. 

This all makes we wonder which GPU gets used for hardware scaling  when there are multiple cards, and if there are other tricks in this kind of environment.

directwindow-and-multiple-displays

I'm in 4.6 and if I drag a jit.window onto a specific monitor, the fullscreen command sent to it will reliably target that monitor/gpu with all the expected benefits.

In my multi-monitor patches, I use jit.displays to see how many monitors I have when the patcher loads. Then I tell each jit.window to live somewhere within the coordinates of each monitor. When I send the fullscreen message to all jit.windows, they all go to their proper monitor.

This works for "direct to window" mode with great results.

My experience comes from this, which you are free to copy/modify/critique/ignore (CC license):

http://www.zachpoff.com/site/software/software.html#multiscreener

(It's a Jitter app called MultiScreener that plays multichannel video in sync on multiple networked computers or just multiple monitors on one computer. Source patches included.)

My experience comes from this, which you are free to copy/modify/critique/ignore (CC license):
http://www.zachpoff.com/site/software/software.html#multiscreener

(It's a Jitter app called MultiScreener that plays multichannel video in sync on multiple networked computers or just multiple monitors on one computer. Source patches included.)
-Zach


directwindow and multiple displays