Compiling VASP 4.6 for 64-bit Xeon (x86_64 EM64T) cluster using Intel compiler and MKL

Questions regarding the compilation of VASP on various platforms: hardware, compilers and libraries, etc.

Moderators: Global Moderator, Moderator

Post Reply
Message
Author
User avatar
Mahesh_chandran
Newbie
Newbie
Posts: 8
Joined: Wed Jun 15, 2005 11:27 am
Location: Bangalore, INDIA.

Compiling VASP 4.6 for 64-bit Xeon (x86_64 EM64T) cluster using Intel compiler and MKL

#1 Post by Mahesh_chandran » Thu Nov 24, 2005 9:42 am

Dear All,
I am planning to compile VASP 4.6 for a cluster of 64-bit Xeon processors (x86_64 EM64T) using Intel compiler (v.9.0) and Intel MKL (v.8.0). The MPICH has been build using Intel compiler. If anyone has answers to the questions below, that would be of help to us:

(a) What flags and libs need to be used in the makefile? We will appreciate if the makefile can be send/posted.

(b) The MKL (v.8) contains files for EM64T. Which of the files need to be linked and how good is the preformace with these libraries?

Thanks
Mahesh
Last edited by Mahesh_chandran on Thu Nov 24, 2005 9:42 am, edited 1 time in total.

tjf
Full Member
Full Member
Posts: 107
Joined: Wed Aug 10, 2005 1:30 pm
Location: Leiden, Netherlands

Compiling VASP 4.6 for 64-bit Xeon (x86_64 EM64T) cluster using Intel compiler and MKL

#2 Post by tjf » Thu Nov 24, 2005 12:25 pm

It's pretty painless: just jump in!

Base you build on one of the ifc makefiles. Just change -xW to -xP and put in the correct library paths. I also have the optimisation turned down to -O2 on pseudo.o (though I can't remember quite why I did that).

I've never had much success with MKL (in terms of imcreased performance), so I use ATLAS etc. with no problems.

There is one problem I've had. VASP+ifort+MPICH+PBS don't play well together on our cluster. About 20% of jobs with that combination killed pbs_mom on the compute nodes, which requires sysadmin intervention. Removing stderr/stdout joining reduced that to about 5%, but still a pain. Switching to OpenMPI (www.open-mpi.org) and avoiding PBS's TM mechnism has stopped the problem and runs beautifully. (I highly recommend OpenMPI in general.)
Last edited by tjf on Thu Nov 24, 2005 12:25 pm, edited 1 time in total.

midair77
Newbie
Newbie
Posts: 11
Joined: Mon Apr 02, 2007 11:32 pm

Compiling VASP 4.6 for 64-bit Xeon (x86_64 EM64T) cluster using Intel compiler and MKL

#3 Post by midair77 » Thu May 10, 2007 1:47 am

Hi, tjf. I am also in the process of compiling VASP to work with openMPI but I have had all kinds of problems.

Her e is the link to my problem on openmpi mailing list: http://www.open-mpi.org/community/lists ... 5/3214.php

Could you kindly describe the steps of how you got openmpi configured like configure options? what Fortran and C compilers you used? What you changed in VASP's make files? How you run your jobs in PBS script?

Besides me, there are other people who also have had problem of getting VASP to work with OpenMPI so your instructions would be a great helps to a lot of people.

My environment: Intel Xeon 5130, Torque + Maui, Intel Fortran 9.1.043 and MKL 9.0/8.1, Gnu gcc and g++ 3.4.6, ACML 3.6.0 on CentOS x86_64. I am trying OpenMPI 1.2.1.

Thank you very much.
Last edited by midair77 on Thu May 10, 2007 1:47 am, edited 1 time in total.

Post Reply