HPL Benchmarking

Discussion created by ssawyer on Aug 14, 2018
Latest reply on Sep 8, 2018 by ssawyer



I am trying to run HPL on an AMD Epyc node (dual socket). I'm getting pretty low numbers so I wanted to see if anyone has had success getting 80-90% of theoretical peak with HPL, i.e. what compiler, MPI version, OMP settings, HPL config , BIOS settings, etc.


I've tried a variety of configurations in HPL.dat including things like 30000 to 82000 for N, 192/200/212 for NB, and a variety of P's and Q's.


I compiled the latest HPL and BLIS framework using version 6.3.0 of the Gnu compiler collection.