Scrubbing Speed

[user_deleted]

Working with a human rig, Osipa-style, with a blendshape type plugin called DPK-BCS.

When I move my controls to deform the face, Maya is lightning fast. At least 70 frames per second.

But when I scrub the timeline, it's another story. I'm down to maybe 5 frames per second. If I'm going to be doing a lot of animation (and I am), that's really a drag.

I've tried deleting absolutely everything in my scene that I don't need. Uneeded textures, uneeded objects, audio clips, extraneous animation controls, etc etc. I'm basically left with just the head geometry, and the Osipa-style facial controls. And STILL, when I scrub, I'm at 5 fps. And still, when I move my animation controls, I'm at 70 fps.

So I know it's not the skinning deformation math that's slow, since playing with the controls is so fast. It follows that it's not the graphics card either.

I really can't figure it out, and this is something I've been more or less living with for months, but now I'm at the point where I really need to start animating a lot of shots, and it needs to be more interactive than this.

Any suggestions? Thanks for any thoughts.

[joojaa]

hmmm its almost impossible to say offhand. But it certainly sounds like you have some extra history or expression slowing things down.

See if you scrub a value maya evaluates only the minimum shard tof the network but if you change time maya changes all things time affects, and its rather easy to build rigs so thetthe solver goes deeper into the tree than necessary.

look at your history tree and see what all is affected by time and hint each one down and time their performance.

[user_deleted]

QUOTE
look at your history tree and see what all is affected by time and hint each one down and time their performance.

What would be your preferred editor/method for doing this? Namely, how do I know what is affected by time, or hunt down the performance of a given node?

My method right now is more midieval, ie. slash things from the scene to see if they speed up... but that hasn't worked.

[jacquesbroquard]

Hi,
You may want to try this script, it plays your keys back in real-time:
abxPlayKeys Script
Best,
Jacques

[joojaa]

Well, the rig is just programming language no one says you cant do it the stupid way around.

One of the classic examples for very stupid hookup is the sparate head to run blend on it and then reattach it, its perfectly valid but it uses up 4 times the time needed for the effect with more memory.

Same applies to acceleration structures, so for example taking head off in hopes of speeding things up does NOT help you you detach the head just in the end of the pipeline. It just speeds the viewpoints but thats rather seldom the issue. Its far better to build a equally rigged no hoistry proxy thst concurrent by the side of it. and then jsut swap the models visibilities.

Can you post a snapshot of your history graph?

[user_deleted]

QUOTE
Can you post a snapshot of your history graph?

Do you mean my hypergraph tree?
If so it looks like this, but it's unreadable at this scale. If this is what you meant, let me know and I can tile together a larger image.

What do you mean by
QUOTE
Its far better to build a equally rigged no hoistry proxy thst concurrent by the side of it. and then jsut swap the models visibilities.

Particularly what do you men by "hoistry", and would you hook up this model to your animation controls also?

[joojaa]

No that shows the hierarchy i want to see the history! so choose graph selected.

[user_deleted]

Okay it looks like I solved it. I found a number of unused image planes in the HyperShade that were referencing non-existent image sequences... so every frame I scrubbed, it was doing disk access looking for files that didn't exist. Must be a veery slow process.

The scene is much much faster now that I've deleted them.
Thanks for being willing to help out... again.

[joojaa]

QUOTE(holographix @ 10/22/07, 05:58 AM) [snapback]275192[/snapback]
Must be a veery slow process.

every time you fetch one thing up on disk ist like taking a stop to count to 1000000, every time the computer accesses memory its liek taking a stop to count for 100, so if it fits cache snugly its way faster.