tag:blogger.com,1999:blog-5188121942175615593.post1619345682088554557..comments2024-03-29T01:51:12.527+11:00Comments on Lindqvist -- a blog about Linux and Science. Mostly.: 445. GAMESS US 2013 R1 on Debian (Wheezy) -- w/o GPUlindqvisthttp://www.blogger.com/profile/11525548328692436184noreply@blogger.comBlogger8125tag:blogger.com,1999:blog-5188121942175615593.post-86423886310532182562014-02-05T22:11:55.703+11:002014-02-05T22:11:55.703+11:00I should add that splitting a job across nodes oft...I should add that splitting a job across nodes often incurs a heavy performance penalty, depending on how they are interconnected (e.g. inifinibad vs gigabit ethernet and whether you use e.g. NFS) so it might not be worth the hassle.lindqvisthttps://www.blogger.com/profile/11525548328692436184noreply@blogger.comtag:blogger.com,1999:blog-5188121942175615593.post-4680098087297380202014-02-05T22:10:11.740+11:002014-02-05T22:10:11.740+11:001.1 Are you using ATLAS? Have you installed ATLAS?...1.1 Are you using ATLAS? Have you installed ATLAS? Note also that this was written for debian -- canonical has a way of doing weird things with libraries for no discernable reason e.g. ECCE won't compile on the latest versions of ubuntu but works just fine on debian and fedora derivatives.<br />Either way, to figure out where e.g. libatlas.so.3 is do<br />locate libatlas|grep so<br /><br />1.2 The line is correct, and that's for openmpi. Again, make sure you've install openmpi and that the location is the same under ubuntu as under debian. Do e.g.<br />locate libmpi.so<br /><br />If LD_LIBRARY_PATH isn't set you can try shortening the line to<br />setenv LD_LIBRARY_PATH /usr/lib/openmpi/lib<br /><br />2. I'm not sure whether gamess is node-aware. I've basically never done any production work with it. However, you can set up openmpi to split jobs across nodes as it sees fit. See e.g. http://www.open-mpi.org/faq/?category=running#mpirun-scheduling<br /><br />Hope that it helps<br />lindqvisthttps://www.blogger.com/profile/11525548328692436184noreply@blogger.comtag:blogger.com,1999:blog-5188121942175615593.post-75270257673054961292014-01-30T22:02:13.998+11:002014-01-30T22:02:13.998+11:00Hi,
Thank you so much for all these topics. I am v...Hi,<br />Thank you so much for all these topics. I am very-very basic with linux command since I am a pure chemist... but trying to develop calculations.<br /><br />Thanks to you, I was able to compile gamess with opempi and atlas. This topic work on ubuntu 13.10. My machine is a HP z600 (2x4 cores, hypertreaded --> 16 cores with 24Gb) and calculations work fine. I have two similr machines, each contains a compiled version according to your topic.<br /><br />But now I want to go further and I am trying to built cluster using kestrelHPC (the only one where a full topic help me to buit one) but in this case only ubuntu lucid works or debian 6.0.x.. I install it on a third computer as frontend machine. <br />I choose lucid in which the gfortran version is 4.4. Almost everything is ok during compilation. Kestrehpc is correctly installed : ganglia returns me 2 nodes (32processes and 48Gb).<br /><br />I have 2 problems for gamess compilation and 1 question about cluster:<br /><br />1.1 problem1 <br />the symlink you put in the topic doesn't work (file not found). Effectively, I looked at the directory and no such file, even not with similar name (libatlas.so.3 and libf77blas.so.3). Do I need to symlink something ?<br /><br />1.2. problem 2<br />in your gmrun, i seem that the line :<br />setenv LD_LIBRARY_PATH /usr/lib/openmpi/lib:$LD_LIBRARY_PATH<br />is not correct. message is that the variable LD_LIBRARY_PATH is not set and program cannot find library. Which library ? altas or mpi ? the folder /usr/lib/openmpi/lib exist on my ubuntu lucid.<br /><br />2. question : about cluster and compilation<br />do I need to specify somewhere the presence of nodes or if I run directly with 32 processes (ie gmrun test.inp 13r1 32 >test.out), the gamess/opempi will automatically send to nodes ?<br /><br />Thank you in advance for your help/comments. if you need to contact me, use phaubert2@wanadoo.fr to email me.<br /><br />sincerly,<br />phAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-5188121942175615593.post-74406523212616248392013-11-18T14:22:01.183+11:002013-11-18T14:22:01.183+11:00I don't have any plans on compiling with open6...I don't have any plans on compiling with open64. I used open64 a few years back to try to compile nwchem, and had little luck. If you do sort it out and write it up, feel free to post a link here.lindqvisthttps://www.blogger.com/profile/11525548328692436184noreply@blogger.comtag:blogger.com,1999:blog-5188121942175615593.post-39412774776413678242013-11-17T23:19:36.514+11:002013-11-17T23:19:36.514+11:00Hi I am thinking GAMESS-US compiling with OPEN64 c...Hi I am thinking GAMESS-US compiling with OPEN64 compiler for best performance at AMD CPUs.<br />Do you have a plan of GAMESS compiling with OPEN64 compiler ?Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-5188121942175615593.post-15088971761303235462013-11-17T23:10:16.419+11:002013-11-17T23:10:16.419+11:00Hi I was tested the performances of GAMESS-US usin...Hi I was tested the performances of GAMESS-US using various compilers(ifort, gfortran, pgfortran). but the OPEN64 compiler was not tested because no support of GAMESS-US.<br />The OPEN64 compiler was best performance about 100% for Gaussian03 at the AMD CPU computers in my experiences.<br />Do you have a plan to compile the GAMESS-US with OPEN64 compiler ?<br />I am now editing the compile scripts and testing the result having a lot of problems.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-5188121942175615593.post-54788731491526448582013-06-27T09:28:04.552+10:002013-06-27T09:28:04.552+10:00Thank you for the link (and the series of posts!)....Thank you for the link (and the series of posts!). <br /><br />I'll try following the instructions next week to see if I have any more luck with libcchem. <br /><br />I added a comment to http://biochemicalmatters.blogspot.com.au/2013/06/gamess-us-frequently-asked-questions_26.htm pointing out potential issues with gfortran 4.7, the optimization level, and test job exam44.inp.<br /><br />I'd be interested in seeing the GUI. To date I'm only aware of http://www.chemsoft.ch/qc/gameix.htm<br /><br />Although to be fair my 'dream' would be for ECCE to support GAMESS US -- ECCE handles job preparation, submission, monitoring, and everything in between. Unfortunately, it <br />1) 'only' handles g03, g98, nwchem, amica and gamess UK<br />and <br />2) the scripts that prepares the input and reads the output from different software packages are written in Perl, and while I find regex as compact and elegant as the next person, I have a hard time reading and understanding Perl.lindqvisthttps://www.blogger.com/profile/11525548328692436184noreply@blogger.comtag:blogger.com,1999:blog-5188121942175615593.post-30785499179354470822013-06-27T02:21:53.453+10:002013-06-27T02:21:53.453+10:00Dear lindqvist,
Today my GAMESS (US) solutions wa...Dear lindqvist,<br /><br />Today my GAMESS (US) solutions was published at http://biochemicalmatters.blogspot.ru/<br /><br />If you interested, please welcome! You can publish my threads, if you want :) Also, I can send you GUI for GAMESS (US) and GUI input file generator, writed on Java.Anonymoushttps://www.blogger.com/profile/01688441971853180344noreply@blogger.com