Thanks. This is strange as I still get 15-20% consistently better results in favor of older versions on three different machines already. Two more studies on other systems attached below.
TEST: Step-23 (integration time modified from 5 to 150, output suppressed) CMAKE_BUILD_TYPE: "Release". MACHINE 1: Intel(R) Core(TM) i7-6700K CPU @ 4.00GHz $ cat /etc/redhat-release CentOS Linux release 7.4.1708 (Core) $ gcc --version gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-16) deal v8.1.0 (built and installed from source): $ time ./step-23 real 1m23.768s user 5m46.080s sys 0m4.079s deal v8.5.1 (built and installed from source): $ time ./step-23 real 1m42.416s user 5m37.018s sys 0m4.340s MACHINE 2: Intel(R) Xeon(R) CPU X5690 @ 3.47GHz $ lsb_release -a Description: Ubuntu 14.04.5 LTS $ gcc --version gcc (Ubuntu 4.8.4-2ubuntu1~14.04.3) 4.8.4 deal v8.1.0 (built and installed from source): $ time ./step-23 real 2m49.114s user 11m41.429s sys 0m48.882s deal v8.5.1 (built and installed from source): $ time ./step-23 real 3m20.583s user 10m54.850s sys 2m18.989s On Tuesday, December 26, 2017 at 6:08:17 PM UTC+3, Matthias Maier wrote: > > Hi, > > I get relatively comparable results for both versions: > > dev: ./step-23 55.55s user 1.64s system 131% cpu 43.637 total > 8.0: ./step-23 55.85s user 1.48s system 129% cpu 44.130 total > > Is this the unmodified step-23 tutorial program? > > For measuring performance regressions a total runtime of less than 5 > seconds doesn't say that much. Never versions allocate and precompute > quite a bunch of stuff upfront which might result in a small (problem > independent) fixed runtime overhead (of a second or less). > > Best, > Matthias > > > > On Tue, Dec 26, 2017, at 05:52 CST, [email protected] <javascript:> > wrote: > > > Deal.II 8.5.1 seems to be 20% slower than 8.0.0. This is the timing I > get > > when running the Step-23 tutorial (output to screen and vtk is > suppressed): > > > > deal.II version 8.0.0: > > > > $ time ./step-23 > > Number of active cells: 16384 > > Number of degrees of freedom: 16641 > > > > real 0m3.432s > > user 0m6.320s > > sys 0m0.612s > > > > deal.II version 8.5.1: > > > > $ time ./step-23 > > Number of active cells: 16384 > > Number of degrees of freedom: 16641 > > > > real 0m4.430s > > user 0m7.080s > > sys 0m0.492s > > > > In general, I get about 20% slow down for my own code when upgrading > from > > 8.0.0 to 8.5.1. What is the reason of such a slow down? Does the deal.II > > follow the right direction given new versions become gradually slower?! > -- The deal.II project is located at http://www.dealii.org/ For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en --- You received this message because you are subscribed to the Google Groups "deal.II User Group" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
