Hi Francesco, in order to have performance reproducibility be sure to: -disable hyperthreading -disable node health check mechanism In both cases I experienced a slowdown up to a factor 2. You also mention that you see slowdown while using threads. The system you mention looks way to small to experience any significant benefit from MPI+OpenMP execution.
BR, Nicola On 11/23/2016 12:09 PM, Francesco Pelizza wrote: > Hi Dear community, > > > I have a question...Since the qe 6.0 release i started to use it, and I > noticed a slow down for systems up to 48 atoms / 100 electrons running > on few cores, and a speed up running upon more cores. > > I other words, taking as example an insulator polymer, set in its > lattice with 96 electrons: > > using qe 5.4 on 8 threads takes 25-35% less time than qe 6.0 > > that's generally true from scf, to vc-relax to bands and phonon or > whatever calculations > > if I scale on servers or HPC I do not see slow down, and perhaps the qe > 6.0 is in the average 10-15% faster. > > > Was it expected to be so? > > Something changed in the way the system is fragmented across threads? > > > BW > > Francesco Pelizza > > Strathclyde University > > _______________________________________________ > Pw_forum mailing list > Pw_forum@pwscf.org > http://pwscf.org/mailman/listinfo/pw_forum -- Nicola Varini, PhD Scientific IT and Application Support (SCITAS) Theory and simulation of materials (THEOS) ME B2 464 (Bâtiment ME) Station 1 CH-1015 Lausanne +41 21 69 31332 http://scitas.epfl.ch Nicola Varini _______________________________________________ Pw_forum mailing list Pw_forum@pwscf.org http://pwscf.org/mailman/listinfo/pw_forum