From dyang@acs.ucalgary.ca Tue Aug 16 00:32:47 1994 Received: from acs7.acs.ucalgary.ca for dyang@acs.ucalgary.ca by www.ccl.net (8.6.9/930601.1506) id AAA16304; Tue, 16 Aug 1994 00:02:17 -0400 Received: from acs2.acs.ucalgary.ca by acs7.acs.ucalgary.ca (AIX 3.2/UCB 5.64/4.03) id AA15152; Mon, 15 Aug 1994 22:11:08 -0600 Received: by acs2.acs.ucalgary.ca (AIX 3.2/UCB 5.64/4.03) id AA19942; Mon, 15 Aug 1994 22:11:07 -0600 Message-Id: <9408160411.AA19942@acs2.acs.ucalgary.ca> Subject: Summary on modeling corrosion inhibitors To: chemistry@ccl.net Date: Mon, 15 Aug 94 22:11:06 MDT From: "Danya Yang" Cc: dsmith@uoft02.utoledo.edu X-Mailer: ELM [version 2.3 PL11y] Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Length: 7112 Many thanks to those who replied to my query about modeling corrosion inhibitors. I've got three replies, and they are all very helpful indeed. Here are my original question and the replies I've got: My original query was: > Dear Netters, > > I am planning to do some modeling work on corrosion inhibitors (organic > compounds) by studying adsorption behavior of a molecular layer at the > metal-solution interface. I am looking for a visualization software which > can show how those organic molecules pack together (to form a molecular > layer) on a metal surface, and how the molecular layer interacts with the > metal surface, or a software which models adsorption phenomenon ( physical > or chemical ). > > I appreciate any suggestions from you regarding to the softwares or published > papers in this field. I'll summarize the replies if there is enough interest. > > ========================================================================= Message 1: from Craig W. Burkhart (burkhart@rds325.goodyear.com) Message-Id: <9408041310.AA12996@rds325> To: "Danya Yang" Subject: Re: CCL:any software for corrosion inhibitors I read with interest your desire to model metal-solution interfaces. As far as the visualization is concerned, just about any of the commercial packages from MSI, Biosym or Tripos will do. AVS from DASGroup would also probably work for these calculations. More importantly, I will pose to you, what methods are you going to use in this study? Using quantum chemical approaches will be tedious and prone to errors in judgment. Using molecular mechanics is okay, but you better be sure that the parameterizations are good--otherwise you will get garbage. My advice is to find a good molecular mechanics forcefield to get a physical feel for the problem. By "good forcefield" I mean that it is well-parameterized for your group of elements in this application. Then, after you've done the "cheap" calculations, do the quantum chemistry on small clusters. The metals-organics problem is one that has been around for some time now, and has had some small successes. My guess is that you will have to borrow some of the techniques from the solid-state physics folks (viz., metal pseudopotentials for the quantum chemistry) to make the problem tractable for modestly sized metal surfaces. >From the molecular statistical mechanics standpoint, you will need to examine the various methods of computing interfacial adsorption isotherms. Computing Langmuir, etc., isotherms can be readily done using molecular dynamics, but like the quantum chemistry, is frought with the quality of the parameterization (that is, the applicability of the forcefield) and size limitations dictated by your hardware. Btw, don't forget that you will also have to master techniques to generate the metal surface. The most obvious, and most difficult (but also the most realistic), technique to generate an amorphous oxide surface (because you WILL have to take a look at them--clean metal surfaces are probably not meaningful. Almost all metals, except for the noble metals, will form an instantaneous oxide surface) is to use stochastic (Monte Carlo) methods of modelbuilding. Juan de Pablo's research group at Wisconsin may be able to help you there. The generation of a meaningful metal oxide surface is almost a research project in itself. >From a practical standpoint, Tom Fabish of Alcoa (fabish_tj@atc.alcoa.com) has done many of these things in the past. You may want to contact him (yes, you can tell him I foisted you upon him :-) ), as he can probably help you avoid some of the many landmines that await you. Good luck. If I can be of any further surface, just give me an e-jingle... -------------------------------------------------------------------------- Craig W. Burkhart, Ph.D. Senior Research Chemist E-mail: cburkhart@goodyear.com The Goodyear Tire & Rubber Co. Fone: 216.796.3163 Research Center Fax: 216.796.3304 142 Goodyear Boulevard Akron, OH 44305 ========================================================================= Message 2: from John Meehan Subject: Re: CCL:any software for corrosion inhibitors In-Reply-To: <9408040641.AA17730@acs2.acs.ucalgary.ca> Message-Id: We are currently working with C.S.(Steven) Sikes (Uni. Soth Alabama) on very similar aspects of inhibition of crystallization. We are primarily looking at Calcium oxalates, carbonates etc but are also interested in scale and corrosion. The procedure we have used is to 1) Grow crystals in the presence or absence of inhibitor 2) Examine the morphology using SEM and light microscopy to determine which crystl plane the inhibitor has its primary mode of action 3) Examine the crystal using Atomic force microscopy at the molecular level to actually SEE deformations in the surface (not yet....but we will!!) 4) Model the interaction Step 4 is performed after optimization of the structures using basic ab initio or semi-empirical methods, and with appropriate charge fitting from known physical measurements. The whole lot is then used in the Quanta package with the Charmm forcefield to allow "docking" of the inhibitor with the surface. To date, the results have been excellent!!!!!!! I hope this helps......please post me a copy of your summary :-) Cheers John ---------------------------------------------- John Meehan O CH2-COOH Department of Biochemistry " / University of Tasmania, HO-P-O-C-COOH Australia | \ HO CH2-COOH PHOSPHOCITRIC ACID ---- A powerful, natural inhibitor of Pathological Biomineralization ---------------------------------------------- ========================================================================= Message 3: from Beverly Bendiksen Subject: Re: CCL:any software for corrosion inhibitors To: Danya Yang In-Reply-To: <9408040641.AA17730@acs2.acs.ucalgary.ca> Message-Id: Dear Danya - Sorry for the delayed reply to your posting, but I've been out of the country for a while. I've done some modeling of corrosion inhibitors, and found that Cerius2 distributed by Molecular Simulations is an excellent package for this type of system. It has a very good crystal builder and variety of forcefields and parameters to choose from. Getting software to visualize inhibitor interactions on the surface is actually quite easy, the difficulty is defining the corrosion inhibition model. I've used metal oxides instead of the pure metal as the substrate, since I think this better represents the surface in an aqueous system. Happy computing! Beverly Bendiksen bbendik@telerama.lm.com Pittsburgh, PA ========================================================================= Thanks again, Craig, John and Beverly. Danya From kongsian@nsrc.nus.sg Tue Aug 16 05:32:51 1994 Received: from aba.nsrc.nus.sg for kongsian@nsrc.nus.sg by www.ccl.net (8.6.9/930601.1506) id FAA18243; Tue, 16 Aug 1994 05:04:39 -0400 Received: by aba.nsrc.nus.sg (5.0/SMI-SVR4) id AA23755; Tue, 16 Aug 94 17:10:33 SST From: kongsian@nsrc.nus.sg (Kong Sian) Message-Id: <9408160910.AA23755@aba.nsrc.nus.sg> Subject: G92: Help with MP2 FREQ calc. To: chemistry@ccl.net Date: Tue, 16 Aug 1994 17:10:32 +0800 (SST) X-Mailer: ELM [version 2.4 PL21] Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Length: 933 I'd like to thank everyone for their advice and suggestions. Sorry if I don't mention each and everyone as there was quite a handful of replies. To summarise, my problem was an MP2 FREQ calculation with 156 basis functions and 2GB of disk space available. G92 insisted on creating a 3.2GB .rwf file. Suggestions are essentially in the following categories. 1. It cannot be done; I need more disk space. 2. Use SCF=DIRECT to not store the integrals. This saved about 400M, but didn't really solve the problem. 3. Use FREQ=NUMER SCF=DIRECT. I tried this and a scratch file of slightly less than MAXDISK was created. This appears to work and I'm at present running the job with this option. 4. Send me the job and I'll run it for you. Thanks! I'll take up the offer if option 3 fails me. I think this list is simply great. Thanks folks! Kong Sian National Supercomputing Research Centre Singapore From cooksj@ttown.apci.com Tue Aug 16 09:32:56 1994 Received: from ttown.apci.com for cooksj@ttown.apci.com by www.ccl.net (8.6.9/930601.1506) id IAA20552; Tue, 16 Aug 1994 08:58:27 -0400 Received: by ttown.apci.com (5.57/Ultrix3.0-C) id AA28703; Tue, 16 Aug 94 09:06:55 -0400 Date: Tue, 16 Aug 94 09:06:55 -0400 From: cooksj@ttown.apci.com (Stephen J. Cook) Message-Id: <9408161306.AA28703@ttown.apci.com> To: d3g359@rahman.pnl.gov Cc: CHEMISTRY@ccl.net In-Reply-To: <9408160013.AA20159@rahman.pnl.gov> (d3g359@rahman.pnl.gov) Subject: Re: CCL:g92 Reply-To: cooksj@ttown.apci.com >>>>> "John Nicholas" == writes: JN> I'd like to talk to anyone who has run gaussian92 on JN> the SGI IP21 (R8000) chip. Thanks, John In fact, I would welcome hearing about *any* experiences running molecular modeling software on SGI's new chip. ********************************************************************* * Steve Cook cooksj@ttown.apci.com * * Air Products and Chemicals, Inc. Tel. (610) 481-2135 * * 7201 Hamilton Blvd. FAX (610) 481-2446 * * Allentown, PA 18195 * * USA * ********************************************************************* * Emacs - the choice of a GNU generation * ********************************************************************* From brichfor@ezmail.ucs.indiana.edu Tue Aug 16 11:32:56 1994 Received: from ezmail.ucs.indiana.edu for brichfor@ezmail.ucs.indiana.edu by www.ccl.net (8.6.9/930601.1506) id LAA24427; Tue, 16 Aug 1994 11:21:12 -0400 Received: by ezmail.ucs.indiana.edu (1.37.109.11/16.2) id AA283460849; Tue, 16 Aug 1994 10:27:29 -0500 Date: Tue, 16 Aug 1994 10:27:29 -0500 (EST) From: brichfor Subject: CCL: Looking for SCI88 To: chemistry@ccl.net Message-Id: Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Hello Netters, Does anyone know where I can get a copy of SCI88? Thank you in advance. Nancy Brichford brichfor@ucs.indiana.edu From friedman@tammy.harvard.edu Tue Aug 16 12:32:57 1994 Received: from tammy.harvard.edu for friedman@tammy.harvard.edu by www.ccl.net (8.6.9/930601.1506) id LAA25021; Tue, 16 Aug 1994 11:49:44 -0400 Received: by tammy.harvard.edu (5.64/10.0) id AA18427; Tue, 16 Aug 94 11:57:27 -0400 Date: Tue, 16 Aug 94 11:57:27 -0400 From: friedman@tammy.harvard.edu (Dawn Friedman) Message-Id: <9408161557.AA18427@tammy.harvard.edu> To: chemistry@ccl.net Subject: that modelling summary I had thought to get away with a couple of postings to this list -- and thank you, Wally, for your posting! -- but it doesn't seem to stem the tide. From now on, anyone who didn't get the summary from the postings or >from previous mass mailings to requestees should just email me, and I will send them a copy as soon as I see their request, rather than waiting to do a mass mailing. I've been doing this for the past week or so, and it turns out to be easier than compiling a mailing list. Faithfully yours, Dawn friedman@tammy.harvard.edu p.s. I'll be at the ACS meeting in DC next week, so there may be some delays. From csimpson@MSI.COM Tue Aug 16 14:32:57 1994 Received: from schizoid.msi.com for csimpson@MSI.COM by www.ccl.net (8.6.9/930601.1506) id OAA27976; Tue, 16 Aug 1994 14:25:00 -0400 Received: from gomez (gomez.msi.com) by schizoid.msi.com (4.1/SMI-4.1) id AA18729; Tue, 16 Aug 94 14:30:43 EDT Received: by gomez (920330.SGI/911001.SGI) for chemistry@ccl.net id AA05764; Tue, 16 Aug 94 14:30:00 -0400 Date: Tue, 16 Aug 1994 14:29:59 -0400 (EDT) From: Charlie Simpson Subject: PC vs. RISC timings To: chemistry@ccl.net In-Reply-To: Message-Id: Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Dear all: Does anyone have timing benchmarks for ab initio and or MM/MD calcs on PCs and RISC type workstations? Second (My apologies ahead of time if this is a really dumb question), what is the difference between a pentium PC and RISC architecture? I say this because I'm considering buying a pentium. However, if the pentium PC goes south because of RISC's, I'll buy one of them. Thanks ahead of time. Regards, Charlie ******************************************************************************* * * * Charlie Simpson, Ph.D. * * Molecular Simulations Inc. E-mail: csimpson@msi.com * * 16 New England Executive Park Phone: 617-229-9800 * * Burlington, MA 01803 Fax: 617-229-9899 * * * ******************************************************************************* From sliu@mastermodel.ps.uci.edu Tue Aug 16 18:33:00 1994 Received: from mastermodel.ps.uci.edu for sliu@mastermodel.ps.uci.edu by www.ccl.net (8.6.9/930601.1506) id SAA01558; Tue, 16 Aug 1994 18:04:08 -0400 Received: from localhost.ps.uci.edu by mastermodel.ps.uci.edu (920330.SGI/5.17) id AA21037; Tue, 16 Aug 94 15:09:32 -0700 Message-Id: <9408162209.AA21037@mastermodel.ps.uci.edu> To: chemistry@ccl.net Subject: Arrhenius Equation Date: Tue, 16 Aug 94 15:09:29 -0800 From: Song Liu Dear Netter: Do you know any commercial package to predict drug stability using Arrhenius Equation? I will summarize all responses. Thanks Song Liu Chemistry UC Irvine From mizan@engin.umich.edu Tue Aug 16 19:33:01 1994 Received: from pla.engin.umich.edu for mizan@engin.umich.edu by www.ccl.net (8.6.9/930601.1506) id SAA01833; Tue, 16 Aug 1994 18:49:18 -0400 Received: (mizan@localhost) by pla.engin.umich.edu (8.6.8/8.6.4) id SAA04888 for CHEMISTRY@ccl.net; Tue, 16 Aug 1994 18:54:48 -0400 Date: Tue, 16 Aug 1994 18:54:48 -0400 From: "Tahmid I. Mizan" Message-Id: <199408162254.SAA04888@pla.engin.umich.edu> To: CHEMISTRY@ccl.net Subject: predictor-corrector/quaternion method Netters, I am using the predictor-corrector/quaternion subroutines from Allen & Tildesley to simulate SPC water. When I use a 1 femtosecond timestep the program runs properly and gives the correct results for energy, pressure g(r) etc. However, when I use a 2 fs timestep the program 'blows up'. Apparently the OH bond lengths become larger and larger and it just blows up. So it appears to be some kind of numerical instability. Has anybody else had a similar experience ? Or is it just some error in my implementation. Is there a fix ? I will summarize and post replies if there is interest. Thanks Tahmid Mizan email: mizan@engin.umich.edu From csimpson@MSI.COM Tue Aug 16 21:33:02 1994 Received: from schizoid.msi.com for csimpson@MSI.COM by www.ccl.net (8.6.9/930601.1506) id UAA02586; Tue, 16 Aug 1994 20:36:29 -0400 Received: from gomez (gomez.msi.com) by schizoid.msi.com (4.1/SMI-4.1) id AA19482; Tue, 16 Aug 94 20:42:11 EDT Received: by gomez (920330.SGI/911001.SGI) for chemistry@ccl.net id AA06475; Tue, 16 Aug 94 20:41:28 -0400 Date: Tue, 16 Aug 1994 20:41:28 -0400 (EDT) From: Charlie Simpson Subject: Summary - PC vs. RISC timings To: "Comp. Chem. List" In-Reply-To: Message-Id: Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII My thanks to all who responded. Sincerely, ******************************************************************************* * * * Charlie Simpson, Ph.D. * * Molecular Simulations Inc. E-mail: csimpson@msi.com * * 16 New England Executive Park Phone: 617-229-9800 * * Burlington, MA 01803 Fax: 617-229-9899 * * * ******************************************************************************* On Tue, 16 Aug 1994, Charlie Simpson wrote: > Dear all: > > Does anyone have timing benchmarks for ab initio and or MM/MD calcs on > PCs and RISC type workstations? > > Second (My apologies ahead of time if this is a really dumb question), > what is the difference between a pentium PC and RISC architecture? I say > this because I'm considering buying a pentium. However, if the pentium > PC goes south because of RISC's, I'll buy one of them. > > Thanks ahead of time. > > Regards, > Charlie > From jxh@biosym.com Tue Aug 16 20:15:32 1994 Date: Tue, 16 Aug 1994 13:02:56 -0700 From: Joerg Hill To: csimpson@MSI.COM Subject: Re: CCL:PC vs. RISC timings Hi, I have a benchmark programme which has been proven to give very similiar timings as quantum chemical calculations would give. It measures mainly floating point performance, but with pieces of real (application) code. I run it on every computer I can get my fingers on. Here is a list of the timings: c c Computer time % Compiler/Option c sec VAX c------------------------------------------------------------------------------- c CRAY Y-MP2E/164 0.00641 28081 cf77 -Zv -Wf -o scalar -o vector c IBM RISC 6000/590 0.008 22500 f77 -O2 c VP 400-EX 0.011 16364 Optimizing and vectorizing compiler c IBM RISC 6000/370 0.015 12000 f77 -O c CRAY 1 0.022 8182 c SG Indigo (R4000) 0.025 7200 f77 -O2 c IBM RISC 6000/350 0.026 6923 f77 -O c IBM RISC 6000/550 0.026 6923 f77 -O c IBM RISC 6000/340 0.032 5625 f77 -O c IBM RISC 6000/530 0.043 4186 f77 -O c IBM RISC 6000/320H 0.044 4091 f77 -O c IBM RISC 6000/320 0.053 3396 f77 -O c IBM 3090 0.059 3051 Optimizing compiler c SG Indigo (R3000) 0.07 2571 -O3 c SG Indigo (R4000) 0.073 2466 f77 c IBM RISC 6000/590 0.075 2400 f77 c Pentium (60 MHz) 0.08 2250 Linux f2c + gcc -O2 c Sun Sparc Sun4 0.09 2000 -O3 and -O4 c Sun Sparc Sun4 0.10 1800 -fast or -O2 c CADMUS FX/1 (i860) 0.10 1800 -O c Sun Sparc 0.10 1800 -O c SG Personal Iris 0.12 1500 f77 -O2 c IBM RISC 6000/370 0.12 1500 f77 c FX/2800 0.13 1385 -O -uniproc c Siemens 7881 0.16 1125 c IBM RISC 6000/550 0.17 1071 xlf c SG Indigo (R3000) 0.18 1000 -sopt c IBM RISC 6000/350 0.18 994 f77 c SG Iris 4D/60T 0.19 947 f77 -O2 c HP 9000/835 0.22 818 f77 -O2 c IBM RISC 6000/340 0.23 796 f77 c ZKI 'System' 0.23 783 c IBM RISC 6000/320H 0.31 584 f77 c IBM RISC 6000/530 0.31 584 f77 c SG Personal Iris 0.35 514 f77 -O1 c ABC 486/33 EISA 0.37 482 MS F77 5.1 (FPi87) c IBM RISC 6000/320 0.38 477 f77 c TOCOM 486/25/isa 0.50 360 MS F77 5.1 (FPi87) c SG Iris 4D/60T 0.53 340 f77 -O1 c HP 9000/825 0.64 281 f77 -O2 c HP 9000/835 0.67 269 f77 -O1 c 386/33 MHz 1.06 170 Copr. MS 4.01 (FPi87) c VAX 785 1.1 164 c Iris W-3120 1.1 164 Copr. c HP 9000/825 1.3 138 f77 -O1 c Robotron K 1840 1.8 100 c VAX 780 1.8 100 c Acer sys-32/20 2.3 78 Copr. MS 4.0 c VAX 750 2.9 62 c VAX 780 3.1 58 c Acer 1100 3.5 51 Copr. MS 4.0 c BESM 6 3.5 51 (*2/3) Optimizing compiler c Acer sys-32/20 4.5 40 Copr. MS 3.2 (8087.lib, $nofloatcalls) c Cyber 172 4.5 40 c EC 1055 5.1 35 Optimizing compiler (OPT=2) c EC 1040 5.1 35 c Acer 1100 6.1 29 Copr. MS 3.2 (8087.lib, $nofloatcalls) c BESM 6 7.0 26 (*2/3) c IBM 370/155 7.5 24 c AT 286/12 MHz 8.5 21 Copr. MS 4.0 (FPi87) c Acer 1100 8.7 21 Copr. MS 3.2 (8087.lib) c Acer 1100 10.4 17 Copr. MS 3.2 (math.lib) c Iris W-3120 12.0 15 c Atari PC3/8 MHz 14.9 12 Copr. MS 4.0 (FPi) c Schneider PC 1640 15. 12 Copr. MS 4.0 (FPi87) c Atari TT/16 MHz 20. 9 Prospero Fortran-77 (Vmmg 2.1) c Schneider PC 1640 21. 9 Copr. PROFORT c Schneider PC 1640 25. 7 Copr. MS 4.0 (FPc) c AT 286/12 MHz 25. 7 MS 4.0 (FPa) c Robotron A 7150 37. 5 Copr. MS 4.0 (FPi87) c PC AT (16 MHz) 46. 4 MS 4.01 (87-Emulation) c Atari 1040 STF 51. 4 Prospero Fortran-77 (Vmmg 2.1) c Atari 1040 STF 67. 3 Prospero Fortran-77 (Vmmg 1.1) c Schneider PC 1640 89. 2 MS 4.0 (FPa) The difference between a Pentium and a RISC processor is mainly floating point performance (and that's what counts in quantum chemical and MD calculations). You see this in the table, all new RISC chips are faster when you compare the same level of optimization. Typical PC applications do not need much floating point power, so manufacturers of PC chips focus on integer performance. Typical workstation applications (as e.g. quantum chemical or MD calculations) need floating point power and most RISC manufacturers design their chips for it. Joerg-R. Hill -------------------------------------------------------------------------------- Dr. Joerg-Ruediger Hill | Every attempt to employ mathematical methods in the Biosym Technologies, Inc. | study of chemical questions must be considered pro- 9685 Scranton Road | foundly irrational and contrary to the spirit of San Diego, CA 92121-3752 | chemistry ... If mathematical analysis should ever USA | hold a prominent place in chemistry - an aberration | which is happily almost impossible - it would Phone (619) 546-5508 | occasion a rapid and widespread degeneration of Fax (619) 458-0136 | that science. E-mail jxh@biosym.com | A. Comte, 1830 -------------------------------------------------------------------------------- The opinions expressed in this message are my personal opinions and no offical statements of Biosym Technologies, Inc. For informations about Biosym products send mail to: rcenter@biosym.com. -------------------------------------------------------------------------------- From ross@cgl.ucsf.edu Tue Aug 16 20:17:06 1994 Date: Tue, 16 Aug 1994 14:06:35 -0700 From: Bill Ross To: csimpson%gomez@MSI.COM Subject: Re: CCL:PC vs. RISC timings Molecular Mechanics/Dynamics Benchmarks The following benchmarks may be of interest, not only for the thrill of watching the price/performance competition, but also for insights into architectures and for clues about what the molecular modelling community might request of designers. It may be useful to construct a set of comp chem benchmarks, including cases such as these along with QM and semiempirical cases. Two cases are considered: a "small," 274-atom solute in a large periodic bath of water molecules and ions; and a "large," 4282-atom molecule in vacuum. For simplicity, both systems are DNA. The code used is Amber 4.0. Although the measurements have not been taken under controlled conditions, the trials that were repeated yielded quite similar results, probably varying by less than 2%. Formal benchmarks would require a 'bare' machine and might well include wallclock times and running multiple copies of a benchmark simultaneously to force paging. In any case, the numbers that follow must be treated as anecdotal and informal. The form of dielectric constant has a surprising effect on performance: using the normal form (dependent on 1/r) exacts as much as a 30% penalty over the "distance-dependent" form (1/r^2) on architectures that have less support for the square root operation (most notably the SGI series and rs6000). Clearly this is one architectural feature that the comp chem community may want to lobby for. Only one parallel architecture was tested as such: an old 8-processor Alliant - using only compiler optimization - obtained correct results on the more tested, older code. The speedup (2.3) is harder to evaluate given the uncontrolled conditions (I don't know how many processors were used). The cases include energy minimization, dynamics, and a free energy calculation. Eventually I expect to run the same cases for the solvated system as for the vacuum one. My thanks to George Seibel and David Case for helpful observations on architectures and factors affecting program speed. Note: The minmd program contains the traditional energy minimization and molecular dynamics capabilities of Amber. Sander is essentially the same, as used here. (Both programs have significant other features which are not exercised by the benchmarks.) Gibbs is the Amber free energy perturbation program. Amber 4.0 Benchmarks These benchmarks are for larger systems than the other demo cases and are intended to compare machine performance on more realistic problems. The order is roughly that of performance for the fastest machine in a product line. All times are CPU seconds measured by system calls in the programs; wallclock times may not correspond. All results except for the Alliant are for a single processor. These are single observations. Note: benchmarks supplied by manufacturers are indicated by a leading '-' in the margin. dna/Run.bench dna/Run.bench2 DNA hexamer in periodic 68 DNA base pairs in vacuum. water box, constant volume. 4282 atoms, 10A cutoff on all 7682 atoms: 274 dna, 10 nonbonded pairs. Distance- counterions, 2466 waters. dependent dielectric. All solute interactions; 8A cutoff otherwise. Constant dielectric. ______________________________ ______________________________ min min+md sander sander gibbs min md ______________________________ ______________________________ Cray C90 - /49 - /56 - /57 - /25 - /25 59 Y-MP(ncsc) - /80 - /92 - /90 - /44 - /44 96 Y-MP(sdsc) - /91 - /104 - /104 - /44 - /44 98 Y-MP EL - /445 - /498 - /500 - /282 - /278 535 Fujitsu VP2200 52/54 62/64 63/64 26/24 26/25 92 HP 735 173/172 197/193 216/190 112/109 109/106 186 730 336/327 367/363 337/363 205/220 200/216 409 720/50MHz 434/462 480/512 476/503 DEC-alpha 3000/500vms 232/275 249/294 247/286 154/191 151/191 258 3000/500osf 285/363 320/397 298/335 160/195 163/197 271 iris** -Chall150.1 209/262 221/277 221/275 153/195 160/191 314 -Chall150.2 117/155 136/170 150/167 104/138 102/128 199 -Chall150.4 88/108 111/124 92/118 79/104 77/104 159 Challenge.1 298/382 322/415 324/405 228/287 228/287 460 Challenge.2 173/219 194/243 193/242 160/196 153/193 298 Challenge.4 108/137 132/182 130/163 118/151 115/147 229 -Challeng.8 79/95 99/120 101/120 -Crimson 308/426 332/455 346/479 -Indigo/R4 310/416 335/448 348/488 Crimson 352/421 341/452 379/468 224/281 226/282 479 w/fastm** 337/416 358/447 325/439 219/273 216/268 481 4d/410vgx 730/1129 779/1180 768/1156 indigo3000 868/1253 881/1391 866/1300 490/629 461/623 1269 4d/310vgx 956/1618 1015/1704 993/1560 578/809 578/804 1244 personal 1722/2830 1724/3371 1724/2846 4d/80gt 1901/3542 1996/4006 2006/3201 rs6000 560 376/347 400/372 405/399 530 859/844 912/895 915/858 516/391 501/378 630 vax 9000 vector 365/468 399/520 390/524 229/311 219/296 no vector 654/774 /865 948/917 462/497 454/794 789 convex c2 479/516 549/597 562/603 279/303 277/304 767 fps 500 744/774 855/865 921/915 mips rc6280 723/1133 758/1191 731/1101 565/869 564/867 888 decstation 5000/200 1112/1585 1173/1657 1168/1638 670/884 663/871 1325 alliant FX/8* 1772/1876 2016/2160 2034/2184 1-process 4270 IBM 3090 200J vector - /1999 - /2051 200J scalar - /6059 - /6143 sun sparc2 1798/2145 1834/2312 1627/2299 sparc 4/280 2528/3830 2700/4062 2708/4018 PROGRAM NOTES Run.bench min: 100 steps minimization minmd: 20 steps min, 80 steps md sander: 100 steps gradual warming Run.bench2 sander/min: 100 steps minimization sander/md: 100 steps gradual warming gibbs: 100 steps of dynamic windows perturbation (double-wide sampling) note: gibbs4 does not have vectorization directives note: gibbs4 is double precision One interesting thing that came to light when developing bench2 was that the distance-dependent (1/r^2) dielectric was significantly faster than the normal (1/r) one. This effect, attributed to the taking of the square root, was more pronounced when hardware arithmetic support was lacking. Representative results (double precision sander minimization): SGI Crim32M MIPS IBM Convex HP Cray Fujitsu diel Crim32M -lfastm RC6280 530 C2 730 Y-MP VP220 1/r^2 347 346 616 391 303 220 44 24 1/r 594 495 869 576 313 240 49 29 ratio .584 .699 .709 .679 .968 .917 .898 .828 When the SGI Crimson32M used -lfastm, the double precision version was faster than the single for the 1/r^2 dielectric: 566 single, 495 double on minimization. MACHINE NOTES The Fujitsu VP2200 is a 32-bit machine with 64-bit arithmetic. The Cray Y-MP is a 64-bit machine, so single precision results are irrelevant. ncsc = North Carolina Supercomputing Center sdsc = San Diego Supercomputing Center The Convex C2 was running under IEEE Floating Point default mode. ----- Cray hpm (hardware performance monitor) results ________ Y-MP ncsc___ ________ C90 ________ Run.bench Run.bench2 Run.bench Run.bench2 MFlops 61.5 72.7 109.0 116.6 MIPS 36.9 37.8 59.0 63.2 M_Mem/sec 70.9 91.2 115.1 142.3 ClockCyc/Inst 4.5 4.4 4.1 3.8 ----- Memory (Mb) Data Cache Instruction Cache Cache c2 1024 fps500 128 alliant 64 512K mips 64 rs6000/530 16 dec5000/200 32 Challenge 1024 16k (100MHz) 16k Sec Cache 1MB Chall150 1024 16k (150MHz) 16k Sec Cache 1MB iris4d/Crim 96 8K 8K Sec Cache 1MB iris4d/indigoR4000 32 8K 8K Sec Cache 1MB iris4d/410vgx 128 64K 64K Sec Cache 1MB iris4d/310vgx 32 64K 64K Sec Cache 256KB irisPersonal 12 32K 64K iris4d/80gt 8 32K 64K *Automatic parallelization directives were invoked in the Alliant compilation. The machine has 8 processors. I do not know what the parallel timings mean, but am impressed that correct results were obtained on all tests except polarization. -Bill Ross ** SGI: Multiprocessor results are from a version of code parallelized by SGI, available on request from UCSF, currently being integrated in the standard release. -lfastm = "fastmath" lib. Energy results were exactly the same after 100 steps min + 100 steps md. -BR -------------------------- testing effect of vector on 4.1 Vector speeds up ~30% on a Convex. 100 steps each test, md1 = gradual warming, md2 = straight dynamics times: novec/vec --dna in water, 7682 atoms-- ---dna in vac, 4282 atoms--- min md1 md2 min md1 md2 Convex x 181/127 196/141 194/142 183/132 181/131 179/114 y 472/320 511/362 514/358 517/338 516/333 511/331 Cray x y ---- Bill Ross >From chemistry-request@ccl.net Mon Jul 25 15:44:54 1994 Received: from cgl.ucsf.EDU (cgl.ucsf.EDU [128.218.14.2]) by socrates.ucsf.EDU (8.6.9/GSC4.24) with ESMTP id PAA10708 for ; Mon, 25 Jul 1994 15:44:52 -0700 Received: from www.ccl.net by cgl.ucsf.EDU (8.6.7/GSC4.24) id PAA23326 for ; Mon, 25 Jul 1994 15:44:48 -0700 Received: for chemistry-request@ccl.net by www.ccl.net (8.6.9/930601.1506) id RAA09781; Mon, 25 Jul 1994 17:58:17 -0400 Received: from agouron.com for hou@agouron.com by www.ccl.net (8.6.9/930601.1506) id RAA09442; Mon, 25 Jul 1994 17:29:45 -0400 Received: from agouron.agouron.com by agouron.com via SMTP (940406.SGI/930901.SGI) for CHEMISTRY@ccl.net id AA15160; Mon, 25 Jul 94 14:34:07 -0700 Received: from localhost by agouron (8.6.4.2/10.3) id OAA22939; Mon, 25 Jul 1994 14:34:04 -0700 Date: Mon, 25 Jul 1994 14:34:04 -0700 From: hou@agouron.ucsf.EDU (Xinjun "Jason" Hou) Message-Id: <199407252134.OAA22939@agouron> To: CHEMISTRY@ccl.net Subject: CCL:Hardware benchmarks Sender: chemistry-request@ccl.net Errors-To: ccl@ccl.net Precedence: bulk Status: R raeker@tc6.fi.ameslab.gov (Todd J. Raeker) writes: > I will be purchasing at least 2 high performance workstations >... >As I write this I realize that most likely these types of questions >come up periodically. Being that most people on this list are ussually >involved in similiar searchs, it might be a good idea to maintain some >sort of FAQ summarizing the real life performance of workstations. If >there is any interest and one does not already exist, I will volunteer >to maintain such a file that would be available via ftp. Let me know >what you think. Several sites contain benchmark information on various hardwares, many benchmarks are not comp. chem. specific: 1) PDS, "The Performance Database Server", http://performance.netlib.org/performance/html/PDStop.html - a lot of information 2) Anonymous ftp: "ftp.nosc.mil" in "pub/aburto" - bechmarks reported by individual users + ... 3) netlib@ornl.gov (gopher: netlib2.cs.utk.edu 70) in benchmarks/performance directory - LINPACK (from the oldest machine to the newest ones) 4) "comp.benchmarks" newsgroup - regular posting of FAQ.benchmarks and other things Xinjun C Xinjun Jason Hou hou@agouron.com C Agouron Pharmaceuticals, Inc. #include C10110000110100101101110011010100111010101101110010010000110111101110101 ---Administrivia: This message is automatically appended by the mail exploder: CHEMISTRY@ccl.net -- everyone | CHEMISTRY-REQUEST@ccl.net -- coordinator MAILSERV@ccl.net: HELP CHEMISTRY | Gopher: www.ccl.net 73 Anon. ftp www.ccl.net | CHEMISTRY-SEARCH@ccl.net -- archive search http://www.ccl.net/chemistry.html | for info send: HELP SEARCH to MAILSERV From apisan@redvax1.dgsca.unam.mx Tue Aug 16 20:17:18 1994 Date: Tue, 16 Aug 1994 15:12:11 -0600 (CST) From: Pisanty Baruch Alejandro-FQ To: Charlie Simpson Subject: Re: CCL:PC vs. RISC timings Dear Charlie: 1. Look into the CCL archives for benchmarks; in a minute I'll send you a relevant file too. 2. Except if you'll only be doing word processing and worksheets, buy a RISC. But MSI had better have someone able to answer your question - if they don't I'll *never* even consider their products again! > >We do have people who can answer this question. I just neglected to ask >anyone. > Yours, Alejandro . . . . . . . . . . . . . . . . . . . . . . . . . . Dr. Alejandro Pisanty, Secretary of the Advisory Council on Computing, UNAM, and Head of the Graduate Division, Faculty of Chemistry, UNAM Universidad Nacional Autonoma de Mexico (UNAM) Ciudad Universitaria 04510 Mexico DF MEXICO Tel. (+52-5) 622 4181, 616 1649; Fax 550 0904, 616 2010 . . . . . . . . . . . . . . . . . . . . . . . . . . From Jie.Yuan@UC.Edu Tue Aug 16 20:17:34 1994 Date: Tue, 16 Aug 1994 20:01:47 -0500 (EST) From: "JieYuan,Chemistry,U.Cincinnati" To: csimpson%gomez@MSI.COM Subject: Re: CCL:PC vs. RISC timings Dear Dr. Simpson: I am a chemist, knowing only enough about computers to get around. But I'd like to share my understanding of RISC vs. CISC with you. These are all hear-says. CISC is developed on the idea of designing complicated (thus the C) instruction system in the CPU chip, so that the programming can be much easier, therefore, the requirement for RAM is much less demanding. It served its purpose in the past decade or so, rather well. Remember RAM's used to cost huge $$? RISC is based on the fact that simplified/reduced (thus the R) instructions can be piped through the CPU and improve the overall efficiency of the CPU very much. It requires a little bit more RAM, but hey, the benefit is huge :-) The so called pipe line can be analoged by an assembly line. The first worker puts on the first, say, 10 parts, and move the product to the second worker. In RISC, the first worker can now pick up the next product and install the first 10 or so parts while the second worker is installing the second 10 parts or so on the first product, simultaneously. In CISC, this is impossible, because the instructions are not the same length and the first worker does not have a way of figuring out what to do untill all the workers finish their parts of a job. In other words, the first worker will have to wait all the time, except at the beginning of a product. So do all the other workers at various times. In a CPU, there are many "parts" of processing, and their work cannot be coordinated well to utilize their potentials efficiently. Their work has to be pipelined to make the whole system efficient. RISC is the future of computing, any efficient computing. At the beginning of the RISC technology, only supercomputers were well pipelines, now most workstations are. In the near future, all the personal computers will be. That is why I like the IBM/Apple/Motorola PowerPC line of CPU. This product has a future, while Pentium does not. Have you heard that the P6 (80686) chip of Intel will be RISC and the 80x86 CISC will be hardware emulated so that all the PC softwares do not become obsolete overnight? CISC's days are numbered. Emulated CISC will put enough overhead on the computing that P6's performance can hardly be better than Pentium's. Only the true RISC software can beat the CISC with ease. However, true RISC PC softwares will take some time to develop. Apple's PMacs are doing it now. PC's will have to do it in 1995 or later, but they cannot avoid it. So, if you have enough money, go for a SGI, Sun, HP, or Alpha, which are all RISC machines. P-Macs are also quite good, not to mention its bang for the buck than any of the workstations, whose profit margins have to be high due to limited volume. Pentium? Probably not, unless you have only the software for PC :-( Cheers! Jie ===== Jie Yuan === Chemistry === U. Cincinnati === Jie.Yuan@UC.EDU =====