6

I'm solving the Helmholtz equation using PETSc. I found with the PETSc configure option --download-f-blas-lapack my program runs twice as fast over running it with MKL. Is this a common trend or are there other factors at play? I'm using gcc and Open MPI with PETSc.

Aron Ahmadia
  • 6,951
  • 4
  • 34
  • 54
Hui Zhang
  • 1,319
  • 7
  • 16

1 Answers1

9

This is usually caused by trying to use a threaded MKL combined with MPI, resulting in over-subscription. Either explicitly configure PETSc to use non-threaded MKL or add MKL_NUM_THREADS=1 to your environment.

Jed Brown
  • 25,650
  • 3
  • 72
  • 130