Unaccounted Frame Time is killing frame rate?

Hi,

I have a project in which many (about 100) images are randomly animated through a fairly complicated set of CHOPs.

For some reason my frame rate has absolutely plummeted, and I can’t figure out if it is the animation network for some reason, or if it is the displaying of the actual images.

When I stop the base COMP that is holding all of the chops from animating, the FPS hits 60, but as soon as I enable it, I hit 12 FPS (even with 3 Quadro k6000s).

Checking the base COMPs Children cook time shows that its children take most of the cook time, although this doesn’t make since because it is just channels being driven by waves.

I ran the performance monitor to see if I could figure out what is slowing down my project, but almost all of the 87 ms frame time is listed as “Unaccounted Frame Time” (see image) / or .txt file.

This is all very weird because I have been using this project for weeks without this issue.

My questions are basically:
A. How can I figure out where the Unaccounted Frame Time is going?

B. In a situation like this, where a base full of chops says its children are taking many ms to cook, could it be that they are just driving something else (the moving of images) and that is what is causing the fps to drop, or does the drop in frame rate necessarily come from the calculations done by the chops themselves.
Put another way, is it possible that the “cook time” on a chop refers to the time and cpu usage it takes for that chop to drive some parameter?

Thanks

Are you using GPU affinity? Using multiple GPUs can cause inconsistent behavior depending on which monitor your TD windows are on. Without GPU affinity a random one of the 3 GPUs will be doing all the work, and the data will be transferred to the other GPUs if they are connected to the monitors showing the actual content. This transfer is very expensive and can greatly affect framerate. The GPU that is doing the work could have changed on you.

Thank you Malcolm,

This does seem possible, however, I tried with Mosaic mode and did not notice much of an improvement in FPS. I also carefully read the the “Using Multiple Graphics Cards” entry on the wiki: Am I correct that Mosaic is the most ideal method to use?

We have actually made a 12 (4x3) mosaic before (I know this is outside of Nvidia’s specs but the Nvidia Control panel does list it is an option) with pretty good performance.

Now I am just trying to run the file with 4 screens and with 3 k6000s, I would think it would be flying.