Render node using only 15% of processor power when rendering with progressive sampler

[kubbek]

I noticed that when rendering via network and using progressive sampler render node starts with 100% processor usage but then drops to 15% (it is 2x8 core xeon machine) and continues like this for hours. Meanwhile master machine renders under 100% load. It works like this in nightly from 17-01-2017. It doesn't happen when using adaptive sampler - it uses 100% of both machines through the whole rendering.
I don't know if anyone noticed such behavior but it happens here regularly.

Best,
Jakub

[vladimir.nedev]

Can you try settings this hidden channel to 80 and see if that helps ?



Greetings,
Vladimir Nedev

[kubbek]

Can you try settings this hidden channel to 80 and see if that helps ?

[ATTACH=CONFIG]35427[/ATTACH]

I changed "progressive noise threshold" to 80 and now the cpu usage on slave is jumping like crazy from 0% to about 60% and back to 0% in less than 10 scecond interval. Looks very strange. Master machine is running very steadily.

Best,
Jakub

[vladimir.nedev]

Or, as someone suggested in the Skype group, you can try increasing the Ray Bundle Size for the Progressive sampler.

https://docs.chaosgroup.com/display/...ressiveSampler

Greetings,
Vladimir Nedev

[kubbek]

Or, as someone suggested in the Skype group, you can try increasing the Ray Bundle Size for the Progressive sampler.

https://docs.chaosgroup.com/display/...ressiveSampler

Greetings,
Vladimir Nedev

I made some test and different Ray Bundle Size causes different behaviour on slave but couldn't find values that work better than defaults

[vladimir.nedev]

I made some test and different Ray Bundle Size causes different behaviour on slave but couldn't find values that work better than defaults

We might need to optimize the DR process when using progressive rendering.
In general, it is a harder task to solve compared to bucket rendering.

Note that how bad or good the DR performance is, depends on your scene.
If you can send us your scene, we can use it to investigate the issue in more depth, now or in the future.

Greetings,
Vladimir Nedev

[kubbek]

We might need to optimize the DR process when using progressive rendering.
In general, it is a harder task to solve compared to bucket rendering.

Note that how bad or good the DR performance is, depends on your scene.
If you can send us your scene, we can use it to investigate the issue in more depth, now or in the future.

Greetings,
Vladimir Nedev

I noticed that the problems with progressive sampler and network rendering are very scene dependent. In the problematic scene I reduced texture sizes so there are now textures bigger than 4K and it made a great impact - now the slave fluctuates between 40-99% cpu usage. Increasing ray bundle size to 512 made things noticably worse - cpu usage on the slave moved from 0% to 100% and back. Looks like ray bundle size of 256 is a sweet spot for this scene but the most impact had textures sizes.