Project

General

Profile

Task #1826

investigate non-optimal results with CPU-GPU balancing on AMD

Added by Szilárd Páll about 4 years ago. Updated 17 days ago.

Status:
New
Priority:
Normal
Category:
mdrun
Target version:
Difficulty:
uncategorized
Close

Description

In some cases the PP-PME balancing increases the cut-off on AMD with OpenCL which turns out to not be optimal. In the attached case the performance is ~10% better with -notunepme.

At the same time, there is also a strangely large GPU wait time; the source of this is likely related to the mis-tuning.

tune_8T.log (22.1 KB) tune_8T.log Szilárd Páll, 09/16/2015 06:31 PM
notune_8T.log (21.2 KB) notune_8T.log Szilárd Páll, 09/16/2015 06:31 PM
topol.tpr (578 KB) topol.tpr water 24k rcut=0.9 nm Szilárd Páll, 09/16/2015 06:31 PM

History

#1 Updated by Szilárd Páll 17 days ago

  • Assignee set to Szilárd Páll
  • Target version set to 2020

Needs to be re-investigated.

Also available in: Atom PDF