[Anouk]

We have an AMD Epyc Workstation running OpenSuse Tumbleweed and installed OpenFoam 5.x & 6.0 both natively compiled (gcc 8.1.1) and a docker version OpenFoam 5.0 & 6.0. Running the same example in pimpleDyMFoam (pimpleFoam in 6) on one core, we got pretty different calculation times

OpenFoam 5.0 docker: 54872s
OpenFoam 5.x native gcc8.1.1: 67117s (+22.3% to docker OF5.0)
OpenFoam 6.0 docker: 98128s (+78.8% to docker OF5.0)
OpenFoam 6.0 native gcc8.1.1: 115184s (+17.4% to docker OF6.0, 71.6% to OF5.x native)

In overall the docker OpenFoam was 17-22% faster than the natively compiled versions.
Is there a good explanation for this. Do the docker versions of OpenFoam have more aggressive compiler settings or is it really an issue between gcc 8.1.1 and gcc 5 Ubuntu 5.4.0 for the docker? If the docker version has a more aggressive compiler setting could this be provided to get also better results in the native compiled versions?

The time difference of Version 5 to 6 is also a little bit annoying but is probably caused by some changes in the standard inner iteration settings that we did not yet figure out. Most likely the PIMPLE iterations termination criteria changed since in OF5 we get 11-12 PIMPLE iterations while for OF6 we get always 20 (maximum number of inner iteration was set to 200).