Troubles with relaxation when using ABINIT 8.8.4

Total energy, DFT+U, BigDFT,...

Moderator: amadon

Troubles with relaxation when using ABINIT 8.8.4

Postby CHILL » Wed Jan 30, 2019 11:55 am

Hello,

I have encountered a problem with my relaxation calculations.
When I run the job using 2018 version of ABINIT (8.8.4) the whole relaxation procedure gives the exact opposite effect: starting from nearly relaxed structure I get something very much worse and eventually the calculations just crashed.
I completed the same job using 2016 version of ABINIT and everything went smoothly.

I've had problems with several different materials, which implies that there can be something different in the newest version of ABINIT.
I have attached the '.out' files for both unstable (2018 version) and stable (2016 version) relaxations.
Could you please tell me if there have been big changes between the versions? Some input that must be added or something else...

Thank you!
Best regards, Mikhail
Attachments
rel_2018.out
The unstable relaxation
(134.09 KiB) Downloaded 13 times
rel_2016.out
The stable relaxation
(310.42 KiB) Downloaded 8 times
CHILL
 
Posts: 18
Joined: Sun Oct 08, 2017 1:37 pm

Re: Troubles with relaxation when using ABINIT 8.8.4

Postby ebousquet » Fri Feb 01, 2019 6:40 pm

Dear CHILL,
Did you compile the two versions with exactly the same compiler version and same compilation flags?
Best wishes,
Eric
ebousquet
 
Posts: 216
Joined: Tue Apr 19, 2011 11:13 am
Location: University of Liege, Belgium

Re: Troubles with relaxation when using ABINIT 8.8.4

Postby CHILL » Sun Feb 03, 2019 4:22 pm

Dear Eric,

Yes, the input was exactly the same. I just changed the ABINIT version in my submit script.
If you want, I can provide all the input!

Best regards, Mikhail
CHILL
 
Posts: 18
Joined: Sun Oct 08, 2017 1:37 pm

Re: Troubles with relaxation when using ABINIT 8.8.4

Postby ebousquet » Mon Feb 04, 2019 10:50 am

Dear Mikhail,
I was asking for the compiler version and the related flgas, in my case I had this problem when compiling with intel17 instead of intel15.
Could you return the output of the following command "abinit -b" for your two different versions (the one working and the one not working?)
Best wishes,
Eric
ebousquet
 
Posts: 216
Joined: Tue Apr 19, 2011 11:13 am
Location: University of Liege, Belgium

Re: Troubles with relaxation when using ABINIT 8.8.4

Postby CHILL » Mon Feb 04, 2019 11:36 am

Dear Eric,

Thank you for your response!
That is the output for the working version:

=== Compiler Suite ===
C compiler : intel16.0
C++ compiler : gnu16.0
Fortran compiler : intel0.0
CFLAGS : -O2 -xHost -ftz -fp-speculation=safe -fp-model source
CXXFLAGS : -O2 -xHost -ftz -fp-speculation=safe -fp-model source
FCFLAGS : -O2 -xHost -ftz -fp-speculation=safe -fp-model source -mcmodel=large -shared-intel
FC_LDFLAGS :

That is the output of the not working version:

=== Compiler Suite ===
C compiler : intel18.0
C++ compiler : gnu18.0
Fortran compiler : intel0.0
CFLAGS : -O2 -xHost -ftz -fp-speculation=safe -fp-model source
CXXFLAGS : -O2 -xHost -ftz -fp-speculation=safe -fp-model source
FCFLAGS : -O2 -xHost -ftz -fp-speculation=safe -fp-model source -mcmodel=large -shared-intel
FC_LDFLAGS :

The compilers are indeed different. Only my versions do not match yours. So the remedy for that would be to change the compiler version?

Best regards, Mikhail
CHILL
 
Posts: 18
Joined: Sun Oct 08, 2017 1:37 pm

Re: Troubles with relaxation when using ABINIT 8.8.4

Postby ebousquet » Mon Feb 04, 2019 1:20 pm

Yes, in the meantime you can re-compile with intel16.
I'm going to ask some Yoda master to see what can be done for this problem...
Best wishes,
Eric
ebousquet
 
Posts: 216
Joined: Tue Apr 19, 2011 11:13 am
Location: University of Liege, Belgium

Re: Troubles with relaxation when using ABINIT 8.8.4

Postby jbeuken » Mon Feb 04, 2019 4:59 pm

Dear Mikhail,

Here is the environment used in our testfarm to test 8.10.2 with intel 18 ( same with intel 15.x/16.x/17.x/19.x )

=== Build Information ===
Version : 8.10.2
Build target : x86_64_linux_intel18.0
Build date : 20190117

=== Compiler Suite ===
C compiler : intel18.0
C++ compiler : intel18.0
Fortran compiler : intel18.0
CFLAGS : -g -O2 -vec-report0
CXXFLAGS : -g -O2 -vec-report0
FCFLAGS : -g -O3 -extend-source -noaltparam -nofpscomp
FC_LDFLAGS : -static-intel -static-libgcc

=== Optimizations ===
Debug level : basic
Optimization level : standard


Our builders run on server with pretty old cpus ( Xeon(R) CPU E5-2623 v4 and Xeon(R) CPU E5-2670 v3 )
On which types of cpu do you run your simulations?

We have noticed for some time now, that there are some "stability" problems in some functions with the new INTEL compilers with the recent cpus... :cry:

We have added a new option to "manage" this problem (which is not always enough :roll: ) :
Code: Select all
--enable-avx-safe-mode  Disable vectorization in problematic procedures

the version of mkl library is also problematic sometimes :o

I'd suggest you at first test :
- replaced mkl by netlib as linalg flavour
- use --enable-avx-safe-mode

Best Regards

jmb
User avatar
jbeuken
 
Posts: 303
Joined: Tue Aug 18, 2009 9:24 pm

Re: Troubles with relaxation when using ABINIT 8.8.4

Postby CHILL » Fri Feb 15, 2019 4:57 pm

Dear jbeuken,

I have consulted the responsible person on our cluster.
He ran some tests and the result was still unsuccessful, so now he is interested in more detailed information:

    1. Which combination of Intel compilers + MKL (exact version) do you recommend
    2. Which is your preferred ABINIT environment: which compiler, which libraries etc., i.e. what do you use when publishing results.

Could you please provide us this information?
I also believe that the newest version of ABINIT crashed in this type of calculation (that I described in the first message) as well.

Thank you!
Best regards, Mikhail
CHILL
 
Posts: 18
Joined: Sun Oct 08, 2017 1:37 pm

Re: Troubles with relaxation when using ABINIT 8.8.4

Postby ebousquet » Thu Feb 21, 2019 6:02 pm

dear Mikhail,
In the meantime, did you try to compile with the intel16 as for your 8.8.4, which was working?
Best wishes,
Eric
ebousquet
 
Posts: 216
Joined: Tue Apr 19, 2011 11:13 am
Location: University of Liege, Belgium


Return to Ground state

Who is online

Users browsing this forum: No registered users and 2 guests