Project

General

Profile

Task #2979

Reconsider default 1 OpenMP thread per thread-MPI rank

Added by Berk Hess about 1 year ago. Updated 8 months ago.

Status:
Closed
Priority:
Normal
Assignee:
Category:
mdrun
Target version:
Difficulty:
uncategorized
Close

Description

We always have 1 OpenMP thread per thread-MPI rank by default. This is best at small/medium rank and simple. But with 128 core nodes, this might no longer be the best. It might even limit the system size when we will soon communicate the whole tpr as one chunk of memory.

History

#1 Updated by Paul Bauer 11 months ago

  • Target version changed from 2020-beta1 to 2020-beta2

#2 Updated by Paul Bauer 10 months ago

  • Target version changed from 2020-beta2 to 2020-rc1

bump

#3 Updated by Berk Hess 8 months ago

Tested 140k membrane protein system with H-bond constraints on Skylake with 96 threads and 96 ranks, 1 thread is fastest.

#4 Updated by Paul Bauer 8 months ago

  • Status changed from Accepted to Closed

Berk, can you upload the data for this?

Also available in: Atom PDF