Project

General

Profile

Task #3171

schedule CPU H2D force contribution in separate stream

Added by Szilárd Páll 5 months ago. Updated 3 months ago.

Status:
New
Priority:
Normal
Assignee:
-
Category:
mdrun
Difficulty:
uncategorized
Close

Description

In order to be able to overlap the H2D copy of CPU force contribution with compute kernels running on the GPU, the copy needs to be scheduled in a different stream than the local/nonlocal NB streams where it it done currently. In the current setup, unless PME completes significantly later than the PP tasks, most/all H2D time ends up non-overlapped.


Related issues

Related to GROMACS - Feature #2888: CUDA Update and Constraints moduleClosed
Related to GROMACS - Feature #2817: GPU X/F buffer opsClosed
Related to GROMACS - Feature #2890: GPU Halo ExchangeIn Progress
Related to GROMACS - Task #3170: investigate GPU f buffer ops use casesClosed

History

#1 Updated by Szilárd Páll 5 months ago

  • Related to Feature #2888: CUDA Update and Constraints module added

#2 Updated by Szilárd Páll 5 months ago

#3 Updated by Szilárd Páll 5 months ago

#4 Updated by Szilárd Páll 5 months ago

  • Related to Task #3170: investigate GPU f buffer ops use cases added

#5 Updated by Szilárd Páll 4 months ago

  • Target version changed from 2020 to 2020-rc1

#6 Updated by Paul Bauer 3 months ago

  • Target version changed from 2020-rc1 to 2021-infrastructure-stable

doesn't seem to be happening

Also available in: Atom PDF