From chemistry-request@ccl.net Tue Jul 13 11:42:52 2004 Received: from celebris04.cetem.gov.br (correio.cetem.gov.br [200.20.105.10]) by server.ccl.net (8.12.8/8.12.8) with ESMTP id i6DGgmFA021079 for ; Tue, 13 Jul 2004 11:42:50 -0500 Received: by CELEBRIS04 with Internet Mail Service (5.5.2655.55) id <36FNMSWD>; Tue, 13 Jul 2004 13:48:18 -0300 Message-ID: <4DB0991764FED5119A2A000475B057CA1C57BD@CELEBRIS04> From: Ian Hovell To: "'chemistry'" Subject: CCL:Summary Choice of computer for G03 Date: Tue, 13 Jul 2004 13:48:17 -0300 MIME-Version: 1.0 X-Mailer: Internet Mail Service (5.5.2655.55) Content-Type: text/plain; charset="iso-8859-1" X-Spam-Status: No, hits=0.0 required=7.5 tests=none autolearn=no version=2.61 X-Spam-Checker-Version: SpamAssassin 2.61 (1.212.2.1-2003-12-09-exp) on servernd.ccl.net As promised here is a summary to the following question >I'm looking for the ideal computer for running G03. I was wondering what >people on the list, when buying a new computer dedicated to Gaussian03, >would go fo? Would people go for cpu speed, cache memory, RAM or hard disk. >Where would the emphasise be? And what OS would you opt for. I'm sure these >questions have been asked for before so if any body can point me to a URL >they know, okay. >Thanks for your help and time. The answers; I don't believe you will get definite answers to your question. I've seen people on this list using computers of all kinds, single or multi processors, Linux, Windows, Mac, etc... running Gaussian. Maybe the most important question is what kind of simulations do you want to do on this computer? Of course, I'd go for the best computer possible, ant that means best CPU, with large cache, maximum RAM and a large HD. Especially, if you plan on doing ab-initio calculations on large molecules, you will need lots of HD space, and maybe consider a 62 bit processor. As for OS, I'd definetely go for Linux, unless you choose a multi-processor computer for which Linux isn't available. HTH, Gustavo Seabra. Ian, I have grown fond of Mac's again, since OS X is the only operating system where I can do both professional development in C and Python alongside Word - all without an emulator and/or dual booting. The prices for dual processors are comparable to what Dell has, if not a little cheaper (list prices, YMMV). Whatever you get, there are a number of things to consider: 1) People forget that there are things like memory bandwidth and disk/IO bandwidth. Thus, they purchase the fastest processor(s) they can, and wonder why their jobs don't run as fast as they'd expected. I'd plan for 1 job per 1 or 2 processors for maximum efficiency. If you have to use integral files, I'd investigate fast disks, SCSI, etc. 2) Compilers make a HUGE difference. gcc/g77 are way slower than machine-specific compilers (Intel/PC or IBM/Mac). While it costs a little $$, getting 20-50% faster run-time is a cheap upgrade. Some compilers also have difficulties with some parallel libraries, or don't offer them - check with Gaussian if there's any good or bad purchases. 3) The safe recommendation is that one should get as much memory and disk as possible. See (1) for kinds and/or speeds of disk. Regarding cache, I don't know how particular portions of Gaussian relate to cache hits/misses - they should know this. Of course, running more than a single job on the box is quite a bit worse re: cache than differences with a single job. And I think people now feel that machines should NEVER page (some turn off the option in Linux), so lots and lots of RAM is required. 4) Do you want to be your own system manager? Do you have access to one who is competent? If not, maybe Linux will be a bother - nothing's real good out of the box as I understand things. I run off-the-shell OS X to success. 5) Do you need any kind of compatibility with PC/Mac/Linux/...? Does the machine have to have a graphics head or is it merely a server - the latter is WAY easier to configure and maintain. I'm not sure how PC's communicate with Mac/Linux/Unix, but the latter three seem to talk to each other well. Hope this helps... Joe I've used various versions of Gaussian on various machines at supercomputing centers for many years (at least 18 years). My current favorite machine is SGI's itantuim/altix machine. It is shared memory so gaussian speed-up is nearly linear (at least up to 8 processors, as many as I have tried), and the 64 bit system makes file size limitations no longer an issue. I have been able to optimizations/frequency calculations on molecules with nearly 150 atoms, including Re atoms, using good quality DFT methods, as well as MP2 calculations that choked other machines I have used. Steve Williams I hope that helps Ian Hovell