I have a scene where MTOA constantly drops to 1 thread interactive rendering.
If I toggle 'auto detect threads' on/off it helps for a moment, all 40 cores are rendering. But as soon as I start tweaking shaders - it again drops to 1 core render.
Any ideas ?
Can you post an Arnold log for an IPR session?
Is the actual bucket rendering that drops to 1, or is it something during node init?
I attach INFO log. You can see arnold is starting less and less bucket workers...
00:04:18 2994MB | starting 37 bucket workers of size 64x64 ...
00:04:34 2994MB | starting 34 bucket workers of size 64x64 ...
00:04:48 2999MB | starting 31 bucket workers of size 64x64 ...
00:05:07 2999MB | starting 28 bucket workers of size 64x64 ...
00:05:27 2999MB | starting 25 bucket workers of size 64x64 ...
00:05:35 2999MB | starting 19 bucket workers of size 64x64 ...
00:05:54 2998MB | starting 10 bucket workers of size 64x64 ...
--- HERE I TOGGLE e 'auto detect threads' on/off
00:06:36 3005MB | starting 37 bucket workers of size 64x64 ...
Did you disable Auto Detect and set a specific number of threads (for example, -4 to use all threads except 4) ?
Yes, I tried setting number of threads to explicit '-4' this does not help.
One think makes me wonder. I exported statistics JSON and I am worried by openImageIO stats (though I don't fully understand them, need to do my homework here). Please find the attached file.
This is the first project where we do not convert textures manually to TX, but instead let maya do the conversion. We then started having warnings for non mipmapped textures, but the render proceeds and uses all cores at first). OpenImageIO_report.zip
I don't think tx generation is the problem, because that happens before bucket rendering.
That log excerpt looks like it is from renders done in Maya. That usually means you will see texture warnings from previous renders (the BROKEN) messages. So I wouldn't worry about that. Just restart Maya, or do a batch render, and then check the texture stats again.
Can you check core usage in the Task Manager while doing IPR? What do you see?
Did you always have this problem? And for how long have you been using MtoA on this machine? A few weeks?
Task Manager show that CPU usage percentage is dropping down from 100%.
For example: I can move a light left and right, and after each move - lesss threads are working.
I've been using this machine for a long time now. I'll try to observe on other scenes/projects and maybe pinpoint what's causing this. I just need some more time to play around.
I was wondering about the usage of individual cores...are there still threads using on the cores that Arnold is not using?
I would probably also test with kick -ipr, to see if it happens with just Arnold, or if it is an Arnold Render View problem.
I don't remember any other cases like this, so there's no previous experience to go by.
Looks like we got exact same issue at our company. Work around that is setting UI threads to "none" in ARV and then in setting set threads to -3 for example.