distributing simulations

[mgt]

Can anyone comment on which is smarter?

to render 8 slices on dual quad core xeons with a total of 8 threads

- OR -

to render 4 slices on each dual quad core xeons with a total of 8 threads but with 2 computers so the total slice number is still 8

I guess what I am asking is, does a slice use more than one thread when baking a simulation?

I am thinking if each slice is assigned dual threads then there will be less overlap (voxel, particle) to be calculated overall.

[lukeiamyourfather]

does a slice use more than one thread when baking a simulation?

Yes, a single fluid simulation (or slice) will use more than a single thread. It won't always use the maximum but it'll use more than one. The distributed simulations are inherently less efficient than a single simulation on a single machine so you want to avoid distributing a simulation that can be reasonably handled by a single machine, and never split a single simulation on a single machine into multiple slices. Also there are artifacts when distributing a simulation which can be pretty bad. Long story short the distributed simulations are a last resort to get big simulations though that wouldn't be able to run on a single machine.

In most cases the low resolution simulation can be done on a single machine without slicing. Ideally only the wavelet upres should be distributed which has little or no artifacts if setup correctly. One machine should be used per slice of the distributed upres, so if you have two machines use two slices. Hope that helps. Cheers!