Maya USD export speed

[bogdanamidzic]

We have scene with around 10 characters, exporting them as USD geometry cache. Noticed that speed of exporting as USD is at least twice slower than Alembic. Only one CPU thread is being used, barely. Are there any plans to improve on this?

I am using regular usdExport since usd layer editor is useless. If you have two geometries with the same name after stripping namespaces, it wouldn't work. So, don't know how nobody noticed that. It is very common to have same names of geometries in proxy and render groups. And we don't want namespaces, they are useless later in the pipeline.

I'm currently using Maya 2024. If speed of exports is improved in newer versions, would gladly switch over.

[bogdanamidzic]

Tested in Maya 2026. Same, pretty slow.

Please, we don't need AI, don't need chatgpt which would copy over boxes into my scene. Things like USD export using single thread barely is something only developers of Maya can solve.

Now for example I'm forced to write deadline plugin which would run several maya instances to export each character in parallel.

[neilrubie]

I would suggest you explore exporting all your characters in a single command using the exportRoots flag. You can then open up the cache after the fact and separate each as needed into their individual caches. Works much faster than exporting each individually.

[bogdanamidzic]

Wasn't exporRoots deprecated? And would that work with strip namespaces?

This is my scene hierarchy:

https://drive.google.com/file/d/100ixZZaFN4KM_S6sO6NC-j9bNpdqcAbh/view?usp=sharing

I want to export only geo groups for each character. It was months ago, but think namespaces were the reason I choose to do each one separately. Will check now

[neilrubie]

To my knowledge I don't think that flag has been deprecated. And yes you're right, namespaces can be a bit of an issue. To get around it, you export WITH namespaces. And during your post process when you separate each cache, you also manually remove the namespace prefix on each primspec.

Of course this all comes with the caveat that all your caches must use the same framerange and samples.

[bogdanamidzic]

Checked, seems namespace is not issue anymore. But the problem to export 2800 frames still stands.
Just to check my sanity, is it possible in Maya to run python commands in separate threads? Idea would be to do exports of smaller chunks of animation and join them in value clip later.
Found this. But seems om.MThreadPool is not working (at least in Maya 2024). There is no class MThreadPool at leas in python.

https://www.dilenshah.com/post/using-the-maya-python-api-to-parallelize-tasks?utm_source=linkedin&utm_medium=blog.post-promoter&utm_campaign=7a6ba74c-8953-4aba-a9d0-6071ae71ecbf

I'm Houdini expert, typical route in Houdini would be to use PDG, to divide frame range in smaller chunks and just run export of these in parallel. If you use local scheduler it would spawn several instances of hython (like mayapy) and grind through it. Your CPU usage will be maximized, of course with cap of available RAM. If you have deadline, you can imagine how much benefits you get. Every task, python script, everything can be distributed to different hyton instances/machines. That is great framework in which people do not have to be c++ gurus and write export plugins from scratch.

Is something like this possible through bifrost maybe?

If not I'll have to go and work on deadline submitter for that.

[wallworm]

Sorry for the long delay. The team was occupied and then the holidays hit.

We will discuss this and get back to you.