Hi there,
I am trying to render a simple VDB sequence in LOPs/Karma, using the Render All Frames With a Single Process option on the USD Render ROP.
The reason I'm using that option is because with it switched off, the Husk process time per frame is about 10x longer than the frame itself takes to render.. so with it switched off the sequence takes ~10hrs to render, with it switched on takes about 1hr.

This technique has worked on other VDB sequences, but for some reason with this particular one I'm having problems, namely that the VDB sequence will get "stuck" at random frames and not progress to the current frame's cache, or suddenly move in world space to a random location and then back again.
If I disable Render All Frames With a Single Process I don't have these problems, but the Husk processing times are a dealbreaker.

I have tried using a VDB sequence on disk and reading it in using a Volume LOP, I have also tried bringing the volume in through a SOP Import LOP, both have the same problem.
The nodes are definitely Time Dependent.

I'm sure I'm missing something simple that has to do with USD time authoring of samples maybe.. any ideas on what I can try to get this to work?

Thanks!

Same question here.

I had issues with doing this when rendering in Karma, and am currently testing with Redshift.
For me it will load and render the first vdb on every frame.

Does the fact that husk needs to load a new vdb on each frame mean that the single process option simply can't be used or is there a trick to it?

same problem here,

"Render All Frames With a Single Process" is problematic in that sometimes the sequence render will get stuck after 2-6 frames unless I enable the "restart delegate" option on the USD rop. This fixes that problem for me but adds a very small time to the overall render time.

If you can share your scene (with the VDB data - or the ability to recreate it), that would be super helpful!

Hi Rob, sorry but this particular example is for a commercial project and under NDA so I can't share it, but when I get some free time I'll try and put together a simple generic test that reproduces it.
Cheers!

I've just run a quick test and can't replicate it again. Which leads me to thinking that in my case the problem was either Redshift or possibly forgetting to correctly cache the LOP network (or that it's a smaller test, or that I'm testing on my M1 mac, or some other thing I haven't thought of).


Assuming the time dependency thing is your issue:
If you have a time dependency then your vdb sequence won't render properly with "Render All Frames With a Single Process".

Setting the frame range on the volume LOP or putting down a cache LOP should do the job.