Register Latest Topics
 
 
 


Reply
  Author   Comment  
ZC

Cooking networks
Registered:
Posts: 142
Reply with quote  #1 

Maya's nucleus solver and fluid solver use up to a few threads to calculate?

Today our company let me test cpu is e5 and i7 parallel computing power. E5 is 32 threads, i7 is 8 threads. The result is that i7 is faster than e5.

If more than a solver to use the most threads, is not the other many threads will not work?
We are not using 8 threads i7, and then the higher the better frequency?

When I use soup nodes on a very large polygon mesh or on a particle, I found that the calculation was very slow and they did not use all of the cpu resources. For example, I used kdTree to make bifrost's wetting map. Use the group node to crop polygons with millions of faces.
Can you make upresFluid, kdTree, point, arrayExpression, and so on, which nodes that need a lot of computation can use more threads? So that they can use cpu of dozens of threads of all resources.

thank peter!!!

0
pshipkov

SOuP Jedi
Registered:
Posts: 4,712
Reply with quote  #2 
Many nodes in SOuP are actually mutithreaded.
I wish it was that easy to multithread everything ...

UpresFluid is multithreaded, so is the kdTree.
Many things in the group node cannot be parallelized because of the nature of the operations performed.
The point and arrayExpression nodes rely on the Maya's scripting engine for parsing and executing of some stuff - that's cannot be multithreaded and there is no way around it unless Maya developers do it for us.
0
ZC

Cooking networks
Registered:
Posts: 142
Reply with quote  #3 

Thank you for your reply.

Can you allow soup's nodes to use more threads?
Not just use 8 threads, but use 32, or 64, or more threads.
Now do the effects of most of the computer cpu use E5, not i7, there will be more threads can be used.

thank peter!!!

0
pshipkov

SOuP Jedi
Registered:
Posts: 4,712
Reply with quote  #4 
Some nodes do. Morph is a good example of that. But adding more threads does not mean faster. You can easily run into thread management overhead and memory bandwidth limitations.
0
Previous Topic | Next Topic
Print
Reply

Quick Navigation: