From owner-nwchem-users@emsl.pnl.gov Mon Jul 16 08:25:44 2007 Received: from odyssey.emsl.pnl.gov (localhost [127.0.0.1]) by odyssey.emsl.pnl.gov (8.13.8/8.13.8) with ESMTP id l6GFPgKC013020 for ; Mon, 16 Jul 2007 08:25:43 -0700 (PDT) Received: (from majordom@localhost) by odyssey.emsl.pnl.gov (8.13.8/8.13.8/Submit) id l6GFPguN013019 for nwchem-users-outgoing-0915; Mon, 16 Jul 2007 08:25:42 -0700 (PDT) X-Authentication-Warning: odyssey.emsl.pnl.gov: majordom set sender to owner-nwchem-users@emsl.pnl.gov using -f X-Ironport-SG: OK_Domains X-Ironport-SBRS: 4.7 X-IronPort-Anti-Spam-Filtered: true X-IronPort-Anti-Spam-Result: AgAAADMqm0aThgBBk2dsb2JhbACPGyIBAQIHBAYHIGg X-IronPort-AV: E=Sophos;i="4.16,542,1175497200"; d="scan'208";a="25928025" Subject: Re: [NWCHEM] NWCHEM in parallel? From: Jozsef Csontos To: nwchem-users@emsl.pnl.gov In-Reply-To: <469B81E6.1040900@psu.edu> References: <469B81E6.1040900@psu.edu> Content-Type: text/plain Content-Transfer-Encoding: 7bit Date: Mon, 16 Jul 2007 10:23:30 -0500 Message-Id: <1184599410.20748.25.camel@bif9.creighton.edu> Mime-Version: 1.0 X-Mailer: Evolution 2.8.3 (2.8.3-2.fc6) Sender: owner-nwchem-users@emsl.pnl.gov Precedence: bulk Hi, I'm almost in the same situation but NWChem was compiled with pgf 6.2 and the performance of the serial code isn't so bad. 1 cpu: 12000 sec; wall 15500 sec 2 cpus: 7000 sec; wall 21000 sec (amd opteron dual-board) Nevertheless, I would prefer to use the parallel version. Are there any suggestions how to improve the parallel performance? Thank you, Jozsef On Mon, 2007-07-16 at 10:34 -0400, Courtney M Krest wrote: > We compiled NWCHEM 5.0 on an intel mac (OS X 10.4.10, 2 processors, 2 > GB) using gfortran. Both the parallel and serial versions run slow. A > job that takes 15 minutes of CPU time takes over an hour of wall clock > time. The parallel version is about twice as fast according to the CPU > clock, but both versions show similar wall clock speeds. > > 2-proc: > Total times cpu: 195.9s wall: 588.4s > > 1-proc: > Total times cpu: 357.0s wall: 593.1s > > What we see is similar to what others have reported in the ARCHIVES in > terms of the CPU usage dropping down to 20-30% for long periods of > time. We are running on a single machine, so the network is not an > issue. Also, serial executables compiled without the USE_MPI=y flag > show similar behavior. Any suggestions? > > > Thanks > > Courtney Krest > > Green Research Group > Department of Chemistry > Pennsylvania State University > 104 Chemistry Building, Box 152 > University Park, PA 16802 > 814-863-0080 -- Jozsef Csontos, Ph.D. (jozsefcsontos_at_creighton.edu) Department of Biomedical Sciences Creighton University, Omaha, NE