Hi Vlad,
You can ignore this. The issue is that the number of grid points adapts to the number of processes being used for load balancing. For higher process counts, more grid points are used and this leads to small numerical changes in the output that are larger for the tolerance of the tests.
In practice, you should adjust the grid spacing to ensure that the required numerical precision is achieved for your application.
Tyler
Sent with Proton Mail secure email.
------- Original Message -------
On Monday, August 1st, 2022 at 8:16 PM, Vlad via AMBER <amber.ambermd.org> wrote:
> Dear Amber Gurus,
>
>
> We came across a test which fails on different platforms using different
> compilers only when number of CPUs > 2, i.e. 4, 8, 16...
>
>
>
> possible FAILURE: check erism.pme.out.dif
> /home/users/isv02/isv0205/amber/parallel/amber20_src/AmberTools/test/rism3d.periodic/1ahoa
> 2c2
> < rism_solventPotentialEnergy -6.1467159933587645E+3
> -3.3545508270242976E+3 -2.7654438137441543E+3 -1.3323764264980484
> -2.5388976163814398E+1
>
> > rism_solventPotentialEnergy -6.1175832282945710E+3
>
> -3.3829653413143600E+3 -2.7070058569739540E+3 -1.5919906150467930
> -2.6020039391210420E+1
> 4c4
> < rism_totalParticlesBox 7.3285706282398201E+2 1.4657112000257143E+3
> 2.7599625187659793E-2 4.0261458766938230
>
> > rism_totalParticlesBox 7.3123564466731000E+2 1.4624670995418790E+3
>
> 3.4423257320050420E-2 4.0329676561506500
> 5c5
> < rism_totalChargeBox -3.9997861302166213 -6.2116964698584172E+2
> 6.2116840710713120E+2 2.7599625187659786E-2 -4.0261458766938221
>
> > rism_totalChargeBox -4.0003200329958740 -6.1979533295506020E+2
>
> 6.1979355732089480E+2 3.4423257320050410E-2 -4.0329676561506490
> ### Maximum absolute error in matching lines = 2.60e-01 at line 2 field 5
> ### Maximum relative error in matching lines = 2.47e-01 at line 4 field 4
>
>
> That is it passes a test for number of CPUs 1 and 2 but fails for more CPUs.
>
>
> The platforms used are Intel (both Gnu and Intel compilers) and A64FX
> (both Gnu and Fujitsu compilers).
>
>
> Thanks,
>
> Vlad
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Aug 04 2022 - 13:37:55 PDT
