Tweet
I'm currently doing some instancing and thought I would try the rayserver instancer. Using it, the translationprocess goes a lot faster but rendertimes take a massive hit (almost 4.5x). Is this to be expected?
-
-
Hi ronald_a , such times are definitely not expected. Can you share your scene here at georgy.chakarov@chaosgroup.com or support@chaosgroup.com -
I've received your e-mail and rendered it. I've got 3-4 sec difference in render time. What are your OS, Houdini and V-Ray versions.Comment
-
win10, h18.0.391, VFH [Info] * Build number:7023 hash:9561d9f from 28 Apr 2020 05:03...
...BUT: I am on a NUMA machine (dual 20 core xeon) which has caused issues (slowdowns) in other areas as well. Could you check on a similar configuration?Comment
-
Well ok, this might be the problem. I'll check and write back.Comment
-
I did another quick test on a macbook pro (8 core) - so non NUMA. I did a new (and very simple)scene with fewer instances (around 28.000) but still see a massive increase in rendertime (0:40 --> 1:20). I'll attach this here as well.Comment
-
Hello ronald_a , I've managed to reproduce the issue and logged it as HOU-1839. Actually it is valid for non NUMA machines also, just the differences are not so big. Thanks for the feedback.Comment
-
Is this something that can be fixed rather quickly (asking because of the current project) or is this something for the core team an will take longer. Would be great to know so I can plan accordingly.Comment
-
1. This is not something that could be fixed easily; by design the rayserver is slower, it should not be that much slower though and I'll investigate this.
2. This is not something that could be fixed easily for NUMA.
3. There is no need to use rayserver for 24K instances, those could be easily rendered without; rayserver should be used for >1M instances which may simply won't fit in memory (on most machines) using "non-rayserver" instancer.Comment
-
This particular case could be fixed though.Comment
-
The 24k instances was just an example scene. My production scene has around 1 million (currently). But this is still rendering fine with the normal instancer. I'm just a little worried what happens when I hit 1.5 or 2 million (won't be a lot more though).
My plan going forward would be to put the instances into a seperate vrscene and reference that back into the main render scene (So I don't get the translation hit every time I submit to deadline and get a bloated render vrscene). Using alembic didn't serve me well there btw.
What do you mean by this particular case can be fixed?Comment
-
> What do you mean by this particular case can be fixed?
This particular case has 2 rayservers, 1 from packed geometry and 1 from copy_to_points; in this particular case I could not export one of the rayservers.
But if you'll have more deep packed hierarchy it'll become slower.
> I'm just a little worried what happens when I hit 1.5 or 2 million (won't be a lot more though).
"non-rayserver" version will just not work (will take dramatically more RAM + much slower compile geometry step).
Comment
-
> This particular case has 2 rayservers, 1 from packed geometry and 1 from copy_to_points; in this particular case I could not export one of the rayservers.
But if you'll have more deep packed hierarchy it'll become slower.
oh, this happened by accident in the example scene. the productionscene should have just on packing level (but I'll check).
> "non-rayserver" version will just not work (will take dramatically more RAM + much slower compile geometry step).
does ram and compile time go up linearly?
Comment
-
I did a test upping the instances to around 2 Million and it still seems to work on the farm machines, so I think I'm fine. Total Frame processing time, including compile and everything goes up a lot, but still lower than using rayserver.
I suppose this will be one for the core team to fix.Comment
Comment