CPU Multi-Threading information help.

[Jlaunois]

Hello, I currently try to simulate a bean drop in a tank farm.


I'm new to tyflow simulation and I have no idea how I can improve my work to have a faster simulation.


The simulation look cool, but need a lot of time to generate the export particle in Tycache mode.


And now I start to be short in time.




When I generate the Tycache files, I look at my computer spec and see that only one of the core of my CPU is 100% used and the other are used to 10% to 20% "sometime". I'm sure that I fail or miss understood how Multi-Treading work.





This is a screen of my tyflow editor and another from my simulation in viewport.


I know there is a lot a particles, but as you can see I need a lot more.




If someone can help, it will be really great.

Thanks

[tyFlow]

From your screenshots, your sim is very heavily reliant on PhysX. PhysX has multi-threading enabled by default, but internally PhysX threading efficiency is related to the number of "islands" that PhysX can split your particles into (islands are isolated groups of particles that can be simulated independently). Since your particles are in a giant pile, you're not going to get much speedup there because they're essentially a giant island that cannot be subdivided - that's why few threads are being utilized here.

In the PhysX settings you can enable CUDA (be sure you install the CUDA DLLs too) to see if that will offer a speedup. That speedup will rely on the power of your GPU for the rigidbody solve.

But in general, a simulation like this is tough to parallelize, simply because you have a huge number of interacting particles that PhysX can't internally separate into non-overlapping groups very easily.

[Jlaunois]

Thanks for your answer.
I will try to creat "islands" and see if my simulation works better.

[tyFlow]

If the particles within the pile aren't going to be moving, you can also convert them to PhysX kinematic ribidgodies with a PhysX Switch operator. They will still act as colliders, but no computations will be required to figure out how to move them around since they'll be static, and it will help to speed up the simulation time.