Allow GPU update without GPU DD

[alexxy/gromacs.git] / docs / release-notes / 2021 / major / performance.rst

diff --git a/docs/release-notes/2021/major/performance.rst b/docs/release-notes/2021/major/performance.rst
index b94c016d48585e286b367c96babdcd8ea25fb55b..85b08653c852b526852e9827ca3248e437134c25 100644 (file)
--- a/docs/release-notes/2021/major/performance.rst
+++ b/docs/release-notes/2021/major/performance.rst
@@ -32,3 +32,10 @@ CPU SIMD accelerated implementation of harmonic bonds
 SIMD acceleration for bonds slightly improves performance for systems
 with H-bonds only constrained or no constraints. This gives a significant
 improvement with multiple time stepping.
+
+Allow offloading GPU update and constraints without direct GPU communication
+""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""
+
+Allow domain-decomposition and separate PME rank parallel runs to offload update and
+constraints to a GPU with CUDA without requiring the (experimental) direct GPU
+communication features to be also enabled.