topic Efficient Core Selection Slowing Down Analysis in Intel® MPI Library

Efficient Core Selection Slowing Down Analysis

EG4 — Thu, 06 Nov 2025 11:16:08 GMT

I run distributed FEA analyses and have upgraded to an I9 13900 which has both P and E cores.

There are 8 P cores and 16 E cores. When I run my analysis, it selects CPUs no. 0,4,8,12,16,20, 24 and 28 as seen in the task manager. It is my understanding that this is a mix of P and E cores. (P cores being 0-16, E being 17-31)

Due to the nature of the analysis, the time is controlled by the slowest core. If an E core is selected, this destroys the analysis speed. My analysis speeds are currently slower than my 2017 processor I upgraded from.

How can I force my analysis to run on the P cores automatically without manually changing the affinity every time?

Apologies if this is the wrong post location.

Cheers

Re: Efficient Core Selection Slowing Down Analysis

Sergey_K_Intel3 — Thu, 06 Nov 2025 15:22:48 GMT

In Intel MPI 2021.15 and newer you can exclude the cores on heterogenous core processors you don't want your MPI application to use:

For example, to exclude the e-cores, you can do:

set I_MPI_PIN=1
set I_MPI_PIN_PROCESSOR_EXCLUDE_LIST=16-31

In the latest Intel MPI 2021.17 you can specify the cores type you want to exclude:

set I_MPI_PIN=1
set I_MPI_PIN_PROCESSOR_EXCLUDE_LIST=ecores

References:

Hope this helps.

Sergey