[Maybe you do not know!] It is about "Time step"

phamad
phamad Altair Community Member
edited May 2022 in Community Q&A

Hi EDEM team and everyone,

I have just realized that the way I select the time step for a simulation has a strong effect on calculation speed.

The case I tested is using the CUDA-based GPU solver with 10 million particles on an RTX A6000 GPU. It is shown that the calculation speed is at least 4 times higher when I use the fixed time step value (I fixed 20%) instead of using "auto time step". And what the GPU Cuda utilization is shown in Task Manager is 60% for the case I used "auto time step", and almost 99% when the fixed time step is used.

I hope this will help :D

Tagged:

Answers

  • Stephen Cole
    Stephen Cole
    Altair Employee
    edited May 2022

    Thanks Phamad, that's useful feedback.  I checked with a smaller simulation file but don't see any difference in speed, we'll check with a larger simulation file like yours and review.

     

    If it's possible to share the simulation file at t=0 without any particles could you provide this for us?  It maybe something specific to the setup or physics models used.

     

    Regards

    Stephen

  • phamad
    phamad Altair Community Member
    edited May 2022
    Stephen Cole_21117 said:

    Thanks Phamad, that's useful feedback.  I checked with a smaller simulation file but don't see any difference in speed, we'll check with a larger simulation file like yours and review.

     

    If it's possible to share the simulation file at t=0 without any particles could you provide this for us?  It maybe something specific to the setup or physics models used.

     

    Regards

    Stephen

    Yes, no problem. Please check my files. 

    The setup is very simple without any specific models used.

     

    Regards, 

    Dzung Pham

  • phamad
    phamad Altair Community Member
    edited May 2022
    Stephen Cole_21117 said:

    Thanks Phamad, that's useful feedback.  I checked with a smaller simulation file but don't see any difference in speed, we'll check with a larger simulation file like yours and review.

     

    If it's possible to share the simulation file at t=0 without any particles could you provide this for us?  It maybe something specific to the setup or physics models used.

     

    Regards

    Stephen

    I also noticed that the difference in the calculation speed disappears when most of the particles are settled down to the bottom (not moving). It means with setups where particles are continuously moving, this difference is very significant! (I attached the files with this setup in this post)

Categories