Hyperthreading / CPU Usage RF 2012

[Pinochio]

Hi all

I´m using RF 2012 on a MacPro 2010, 2x Xeon, 12 real cores, 24 cores hyper threading. RF 2012 don´t use all the cores (24) , see the attached screen copy. The simple scene consist only three circle emitters and a plane .

Why not using all cores ?

Best,
Thomas

[tsn]

Please have a look at the scene's "Simulation options..." (the triangle next to the "Simulate" button). What do you see under "Threads"?

http://support.nextlimit.com/display/rf ... on+Control

Another source for adjusting the threads globally are the preferences:

http://support.nextlimit.com/display/rfdocs/Simulation

RF doesn't always use all cores, because distributing the task to all cores often takes longer than performing the calculation on a single CPU/core. This happens especially with low particle counts/emitter resolutions. You'll see an increase of CPU usage with more particles (or grid fluid domain cells).

[Pinochio]

Thank you Thomas, i´m on the road til this evening. I will test&see again, based on your new information.

Best,
Tom

[Pinochio]

Please have a look at the scene's "Simulation options..." (the triangle next to the "Simulate" button). What do you see under "Threads"?

http://support.nextlimit.com/display/rf ... on+Control

Another source for adjusting the threads globally are the preferences:

http://support.nextlimit.com/display/rfdocs/Simulation

RF doesn't always use all cores, because distributing the task to all cores often takes longer than performing the calculation on a single CPU/core. This happens especially with low particle counts/emitter resolutions. You'll see an increase of CPU usage with more particles (or grid fluid domain cells).

I checked your tips and links. Thanks. I understand the concept now. I played with min/max steps and different amount of threads and observed the different performance of the simulation. So for first precalculations i decrease max steps etc. , test at a first glance the simulation and for the final pass i decrease it. In my small own example scenes i observed only a slightly difference of the performance concerning the amount of threads. I assume in more complex scenes it will be different.

Best,
Tom

[tsn]

With more complex scene you'll definitely see a higher CPU load, but please be careful, because it's not very likely that you'll 100%. RF throughly checks if the task can be distributed to all available cores and full load is not always the fastest solution. It strongly depends on your scene. With meshing, on the other hand, you'll mostly see 90-98% CPU usage.