(manually) Allocating Fluent process to specific cores, avoiding uneven CPU load

    • Rob Vervoort

      Dear all,

      Would anyone know if it possible to assign specific CPU cores to Fluent in batch mode (Windows 10)? I decided to consult this forum since the degree of irritation has reached a certain threshold (and I couldn't find any useful information online)...

      Whenever I start a Fluent process on my machine (e.g. see example batch command below), the question is whether it will run smoothly.
      Most often is does not since the task are not evenly distributed along the 2 CPUs. When e.g. I start a parallel process of 32 tasks, Fluent spawns 24 cores on CPU1/socket 1 (24/24), while only spawning 8 cores on CPU2/socket 2 (8/24). As a result, the calculation is performed inefficiently (much longer calculation time than when tasks are evenly distributed over both CPUs). The current solution is to keep restarting the batch command until the tasks are evenly distributed; but sometimes this takes over 20 restarts, which is not very convenient.

      Example of the batch command that I use:
      start "" "C:/Software/ANSYS Inc/v192/fluent/ntbin/win64/fluent" 3ddp -t32 -g -i case_x_journal.jou

      Would there be a way to assign specific cores of each CPU, or a specific amount of cores per CPU to a Fluent process? Curious to find out if anyone has a solution. Thank you in advance.
      I am running Fluent in Windows (10) batch mode on a 48 core machine (2 x Intel Xeon Gold 6248R).

      Kind regards,

    • Rob
      Ansys Employee

      There are ways to do this, and it's on the OS side. The link we found also contained dire warnings about messing with things best left alone so use with caution. 

Viewing 1 reply thread
  • You must be logged in to reply to this topic.