Displacement presampling using only 1 core

[TheLooopCGI]

Is it just our problem or is the vray displacement presampling not multi-threaded?

The render sometimes take 24 hours on a 16-core server just to finish a medium detailed irradiance map. The CPU usage is around 13% all the time (two cores in use). After the irridiance map prepasses finish the render itself also use between one and two cores.

We can see this behaviour with 2D and also 3D displacement.

Is there something we can do to make the displacement presampling quicker?

Is the amount of tesselation of the object influencing the speed of vray displacement?

[katho3d]

Yes, I have the same problem here on my 8 and 12 core machines. I have to render many scenes with displaced landscapes, so the image calculation is very slow in this regions and the cpu usage jumps between 10 and 50%.

Sometimes the older 8 core machines are even faster than my new 12 core renderslaves. I had to disable Hypertreading in this case to get lower rendertimes. Displacement seems to be a real bottleneck currently.

[vlado]

You might want to increase the "Dynamic memory limit" setting; the default is 500 MB which may be way too low for scenes with a lot of displacement.

Best regards,
Vlado

[TheLooopCGI]

Yea, the default value is quite low in most cases. We used to have it set to 4000 MB and then made it higher to 16000 MB... There is not much difference between 4000 and 16000 though... Either way it's not helping the current problem

[vlado]

Ok then; yep, multithreaded generation is something that we have to work on. Right now, the ray tracing of the displacement is fully multithreaded, but the generation of that displacement is done on a single thread (evaluating the displacement map and so on).

Best regards,
Vlado

[TheLooopCGI]

Ah yes, that explains why the CPU usage goes to proper 100% around the time when render bucket changes position... I guess it calculates something different when it does 100%.

So we look forward to a solution. Thanks.