Re: [gmx-users] short range nonbondeds = 0 in power4
On Fri, 15 Sep 2006, Atte Sillanpää wrote: I've run into a mysterious problem. The versions 3.3. and 3.3.1 compile and execute, but the short range coulomb and LJ energies come out as zero when using the mpi-version. Serial code works ok (mpi version gives zero if run using just one cpu). No errors, no warnings. Hi, the problem seems solved. I removed the --enable-threads from the gromacs config options and now short range forces come out right. Simulations are stable and e.g. water rdfs are identical in power4 and amd opteron. It's a little bit nasty that compiler gets through the threads flag but then computes garbage. It has been said in the mailing list that threads don't work yet, but that's hard to find as the solution unless you know to look for it. To further check my binary I'd like to do some other tests. I tried looking for the benchmarks, but the link doesn't work (search for 'bench' at the gromacs site). Are they still available (and do they have some verified results to compare to?) Thanks, Atte___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php
Re: [gmx-users] short range nonbondeds = 0 in power4
Atte Sillanpää wrote: On Fri, 15 Sep 2006, Atte Sillanpää wrote: I've run into a mysterious problem. The versions 3.3. and 3.3.1 compile and execute, but the short range coulomb and LJ energies come out as zero when using the mpi-version. Serial code works ok (mpi version gives zero if run using just one cpu). No errors, no warnings. Hi, the problem seems solved. I removed the --enable-threads from the gromacs config options and now short range forces come out right. Simulations are stable and e.g. water rdfs are identical in power4 and amd opteron. It's a little bit nasty that compiler gets through the threads flag but then computes garbage. It has been said in the mailing list that threads don't work yet, but that's hard to find as the solution unless you know to look for it. To further check my binary I'd like to do some other tests. I tried looking for the benchmarks, but the link doesn't work (search for 'bench' at the gromacs site). Are they still available (and do they have some verified results to compare to?) Thanks, For now you can download them using good old ftp (user anonymous pass your email) We will fix the website. Atte ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php -- David. David van der Spoel, PhD, Assoc. Prof., Molecular Biophysics group, Dept. of Cell and Molecular Biology, Uppsala University. Husargatan 3, Box 596, 75124 Uppsala, Sweden phone: 46 18 471 4205 fax: 46 18 511 755 [EMAIL PROTECTED] [EMAIL PROTECTED] http://folding.bmc.uu.se ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php
Re: [gmx-users] short range nonbondeds = 0 in power4
Atte Sillanpää wrote: Hi, I've run into a mysterious problem. The versions 3.3. and 3.3.1 compile and execute, but the short range coulomb and LJ energies come out as zero when using the mpi-version. Serial code works ok (mpi version gives zero if run using just one cpu). No errors, no warnings. I've tried using both fftw 2.1.5 and 3.0.1, tried dropping the optimization level to -O1, -qstrict, --disable-ppc-altivec, ... The config options were (e.g.) as follows: setenv CPPFLAGS -I/v/aix52_rs/appl/math/fftw/fftw-3.0.1_32/include setenv LDFLAGS -L/v/aix52_rs/appl/math/fftw/fftw-3.0.1_32/lib setenv MPICC mpcc_r ./configure --prefix=/wrk/gmx --program-suffix=_mpi --enable-mpi --disable-nice --enable-threads -with-fft=fftw3 From the beginning of the config.log: hostname = p690m uname -m = 00105CDA4C00 uname -r = 2 uname -s = AIX uname -v = 5 /usr/bin/uname -p = powerpc /bin/uname -X = unknown /bin/arch = unknown /usr/bin/arch -k = unknown /usr/convex/getsysinfo = unknown hostinfo = unknown /bin/machine = unknown /usr/bin/oslevel = 5.2.0.0 /bin/universe = unknown We don't get the kind of errors described earlier for power4 on the list. I actually managed to make a working mdrun_mpi from 3.3 last year using the same options, but after poe/ppe/... update on the power4 machine I get this odd behaviour also on the 3.3 version. Any ideas? What to tweak, where to look? Cheers, Atte ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php if you suspect PME, then try running without it. -- David. David van der Spoel, PhD, Assoc. Prof., Molecular Biophysics group, Dept. of Cell and Molecular Biology, Uppsala University. Husargatan 3, Box 596, 75124 Uppsala, Sweden phone: 46 18 471 4205 fax: 46 18 511 755 [EMAIL PROTECTED] [EMAIL PROTECTED] http://folding.bmc.uu.se ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php
Re: [gmx-users] short range nonbondeds = 0 in power4
Atte Sillanpää wrote: On Thu, 14 Sep 2006, David van der Spoel wrote: I've run into a mysterious problem. The versions 3.3. and 3.3.1 compile and execute, but the short range coulomb and LJ energies come out as zero when using the mpi-version. Serial code works ok (mpi version gives zero if run using just one cpu). No errors, no warnings. ... if you suspect PME, then try running without it. Hi again, the short range energies come out as zero with all electrostatic methods cut-off, PME, shift. But only the short range. Other components get some numbers (and you _can_ continue the simulation for at least thousands of steps). can you try a small water box on one processor? just to eliminate problems. if you can simulate, that means the forces are correct (or zero too...) try printing the forces in other words. just do one step with all output turned on. Cheers, Atte -- David. David van der Spoel, PhD, Assoc. Prof., Molecular Biophysics group, Dept. of Cell and Molecular Biology, Uppsala University. Husargatan 3, Box 596, 75124 Uppsala, Sweden phone:46 18 471 4205fax: 46 18 511 755 [EMAIL PROTECTED][EMAIL PROTECTED] http://folding.bmc.uu.se ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php -- David. David van der Spoel, PhD, Assoc. Prof., Molecular Biophysics group, Dept. of Cell and Molecular Biology, Uppsala University. Husargatan 3, Box 596, 75124 Uppsala, Sweden phone: 46 18 471 4205 fax: 46 18 511 755 [EMAIL PROTECTED] [EMAIL PROTECTED] http://folding.bmc.uu.se ___ gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-users Please don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php