Firefly and PC GAMESS-related discussion club


 
Learn how to ask questions correctly  
 
 
We are NATO-free zone
 



Re^3: Dependency of SCF convergence on CPU type

Thomas Pijper
thomaspijper@hotmail.com


Dear Yura,

That does look suspicious. Are you sure your Opteron machine is working correctly? A problem with your mainboard, CPU, or memory could be one of the reasons for your convergence problems. I recommend you check your computer with a tool such as Prime95, which puts your computer under heavy load while checking whether the calculations performed give correct answers. In addition, a tool like Memtest86+ can be used to check system memory.

If your computer turns out to be fine, Alex may be able to provide further help (as I'm all out of ideas by then :-) ).


Kind regards,
Thom




On Thu Mar 13 '14 7:59pm, Yura Vishnevskiy wrote
------------------------------------------------
>Hi Thomas,

>>As far as I know, CPU specific versions are only CPU specific because they use certain CPU instructions for speeding up calculations. If you use, for example, a version of Firefly optimized for 'AMD Barcelona/Intel Core 2 Duo and newer' on an old Pentium II CPU you would most likely get an error because Firefly would not be able to access the newer CPU instructions.

>Exactly, I would expect SIGILL or something..

>>SCF convergence should not depend on the CPU you use. The only scenario where it could theoretically matter is when different CPUs or different CPU instructions give extremely small differences behind the decimal mark, and it just so happens that your convergence requires a very large accuracy. But this is a wild guess on my behalf and even when correct probably a one in a billion chance. So, I think a mismatch should not have an effect on convergence.

>I also thought so.

>>Are you experiencing such problems?

>Yes, on one of my machines (Dual-Core AMD Opteron(tm) Processor 2224 SE) Firefly reports

>Warning: this Firefly version was not optimized for this CPU type!
>The overall performance degradation is expected.
>For better performance, use Pentium II/Athlon optimized Firefly version.

>and the SCF cannot converge at all
>

 ITER EX DEM  TOTAL ENERGY      E CHANGE  DENSITY CHANGE     ORB. GRAD      INTEGRALS   SKIPPED
   1  0  0 -2545.611836803 -2545.611836803   0.603671914   0.000000000     1981608795 962944982
          ---------------START SECOND ORDER SCF---------------
   2  1  0 -2549.325014284    -3.713177481   0.209684206   0.051896368     1991213439 967145730
   3  2  0 -2549.422056062    -0.097041778   0.062499986   0.056281921     1984368023 967273806
   4  3  0 -2549.538347014    -0.116290952   0.157502320   0.011598482     1995582920 966711991
   5  4  0 -2549.555705112    -0.017358098   0.203361316   0.005291608     2015929221 965335309
   6  5  0 -2549.556491341    -0.000786230   0.203707492   0.002060392     2006479880 966005299
   7  6  0 -2549.552727162     0.003764179   0.255029945   0.002422195     2015632197 965439724
   8  7  0 -2549.537669888     0.015057274   0.470166764   0.001828124     2024225643 964981627
   9  8  0 -2549.547918792    -0.010248904   0.304085401   0.001784836     2041701351 963766142
  10  9  0 -2549.546948586     0.000970206   0.333240182   0.002313788     2034640656 964114205
  11 10  0 -2549.550300308    -0.003351722   0.340439798   0.004486139     2068864936 962261944
  12 11  0 -2549.553696242    -0.003395934   0.483012604   0.002687016     2025886249 964658436
  13 12  0 -2549.551382539     0.002313703   0.152618014   0.004417314     2053507145 963228192
  14 13  0 -2549.543674104     0.007708435   0.682273715   0.002199642     2040576447 963860799
  15 14  0 -2549.535417303     0.008256801   0.673655593   0.007070052     2093607374 960778251
  16 15  0 -2549.549847626    -0.014430322   0.230508453   0.002900624     2018292070 965330507
  17 16  0 -2549.557520823    -0.007673197   0.140449958   0.002902229     2034227415 964307018
  18 17  0 -2549.549057063     0.008463760   0.233187392   0.002950835     2033758160 964127744
  19 18  0 -2549.553649458    -0.004592395   0.116124454   0.002452793     2021301556 964910486
  20 19  0 -2549.524618269     0.029031189   0.260835181   0.002194053     2019827405 964982886
  21 20  0 -2549.542930606    -0.018312337   0.121940288   0.003030018     2028253860 964544352
  22 21  0 -2549.540965132     0.001965474   0.444849566   0.001672220     2013697408 965424517
  23 22  0 -2549.556756807    -0.015791674   0.254132314   0.003201684     2033178593 964292874
  24 23  0 -2549.543891040     0.012865767   0.198513208   0.002592884     2012943623 965482106
  25 24  0 -2549.564801070    -0.020910030   0.148879661   0.006714372     2022220743 965041853
  26 25  0 -2549.540159131     0.024641939   0.182135306   0.002489723     2010256043 965632092
  27 26  0 -2549.557576372    -0.017417241   0.107437592   0.003246840     2036982250 964158556
  28 27  0 -2549.535438794     0.022137578   0.217099494   0.003551604     2025381109 964714587
  29 28  0 -2549.553726453    -0.018287659   0.204767242   0.003648895     2037511207 964009372
  30 29  0 -2549.550002776     0.003723678   0.254272069   0.002385275     2007904102 965785284

>Same input on the other machine (Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz) does not give the warning and works well.

>

 ITER EX DEM  TOTAL ENERGY      E CHANGE  DENSITY CHANGE     ORB. GRAD      INTEGRALS   SKIPPED
   1  0  0 -2545.611836803 -2545.611836803   0.603671914   0.000000000     1981608795 962944982
          ---------------START SECOND ORDER SCF---------------
   2  1  0 -2549.325014284    -3.713177481   0.209943916   0.051896368     1991213439 967145730
   3  2  0 -2549.437999820    -0.112985536   0.059206228   0.056651146     1984395865 967272956
   4  3  0 -2549.533480397    -0.095480578   0.029888527   0.011549836     1996744755 966594715
   5  4  0 -2549.537196083    -0.003715686   0.012329552   0.005543132     1999207389 966457232
   6  5  0 -2549.539465490    -0.002269407   0.002323627   0.001130723     1999485112 966426083
   7  6  0 -2549.539543678    -0.000078188   0.001769048   0.000596947     1999761031 966403564
   8  7  0 -2549.539569377    -0.000025699   0.000697363   0.000241021     1999890636 966395088
   9  8  0 -2549.539574573    -0.000005196   0.000471781   0.000106257     1999872607 966397833
  10  9  0 -2549.539576316    -0.000001743   0.000254176   0.000052056     1999784540 966402250
  11 10  0 -2549.539576722    -0.000000406   0.000119912   0.000030215     1999750058 966403415
  12 11  0 -2549.539576829    -0.000000107   0.000078273   0.000017797     1999754214 966403066
  13 12  0 -2549.539576863    -0.000000034   0.000036151   0.000011755     1999759124 966402576
  14 13  0 -2549.539576872    -0.000000010   0.000016121   0.000006091     1999766274 966402357
  15 14  0 -2549.539576874    -0.000000002   0.000009388   0.000001836     1999766265 966402399
  16 15  0 -2549.539576875     0.000000000   0.000003445   0.000000726     1999765725 966402444
  17 16  0 -2549.539576875     0.000000000   0.000001246   0.000000358     1999766005 966402441
  18 17  0 -2549.539576875     0.000000000   0.000000676   0.000000134     1999766290 966402433
  19 18  0 -2549.539576875     0.000000000   0.000000269   0.000000063     1999766190 966402431
  20 19  0 -2549.539576875     0.000000000   0.000000142   0.000000032     1999766190 966402431

>I am suing Linux version of Firefly 8.0.0 fully statically linked with MPICH1 (configured for ssh, as I remember).

>Alex, is there a version for Opterons?

>Best wishes,
>Yura


[ Previous ] [ Next ] [ Index ]           Thu Mar 13 '14 10:58pm
[ Reply ] [ Edit ] [ Delete ]           This message read 659 times