Hi,
In the process of numerical simulation using 3DEC 7.0, I have utilized ‘program threads automatic’ to employ multithreading for faster simulation speed. However, the CPU utilization on my workstation is still below 20%. Is there any way to increase CPU utilization?
Thank you in advance!
I’m just a user, but from my experience:
3DEC multithreading used to be based on a distribution of blocks on threads, so that a single block model effectively always ran single-threaded. In such cases, it sometimes helped to split the model, but not too much, since the overhead of the contact calculations could become problematic.
Other than that I never used more than 4 to 8 threads for 3DEC, since after that it even tended to get slower with more threads.
Maybe an official Itascan can give you more guidance, but I suspect some more general information on your model would be required.
Hi，@markus,
Thank you for your response. The computation speed is likely correlated with the number of CPU cores in the computer. My workstation has 28 logical CPU cores, so setting the program to “program threads 28” or “program threads automatic” yields the fastest speeds. Adjusting this number either up or down from there seems to reduce computation speed. Additionally, an official response from Itascan previously mentioned that altering the “block update” parameter could enhance computing speed, yet I observed no significant improvement after making such changes. Given the extensive contact due to DFN in my model, splitting the blocks might not be an ideal solution. Thank you once again for the advice you’ve provided.
Warm regards,
JLUJYP
If you have split the block already with a DFN, there is no need to split them further. As long as you have more blocks than threads, it will do multi-threading.