From chemistry-request@server.ccl.net Wed Dec 22 03:17:53 1999 Received: from lrz.uni-muenchen.de (lsajca1-ar3-012-122.biz.dsl.gtei.net [4.3.12.122]) by server.ccl.net (8.8.7/8.8.7) with ESMTP id DAA23071 for ; Wed, 22 Dec 1999 03:17:52 -0500 Received: (from eugene.leitl@localhost) by lrz.uni-muenchen.de (8.8.8/8.8.8) id AAA06133 for ; Wed, 22 Dec 1999 00:08:36 -0800 Resent-From: Eugene Leitl MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Resent-Message-ID: <14432.34563.163514.834404@lrz.uni-muenchen.de> Resent-Date: Wed, 22 Dec 1999 00:08:35 -0800 (PST) Resent-To: Message-Id: Precedence: bulk From: To: beowulf@beowulf.gsfc.nasa.gov, rgb@phy.duke.edu Subject: Re: BLAS recommendations Date: Tue, 21 Dec 1999 23:28:04 -0500 (EST) From: Emil Briggs > > a) What is the "best" way to add BLAS to a 'wulfish cluster of PPro's >and PII's? I say best instead of fastest or cheapest or GPL'd-est to >allow for a variety of personal interpretations of the word best. My >own would be very much GPL, SRPM based (and fastest possible within that >constraint) if such a thing were possible, but I'd love to hear about >fastest under any circumstances, best for sale, and so forth as well. > b) How is BLAS documented? Is it of the "if you have to ask, you >shouldn't be using it" variety? Books? A manual somewhere? > c) Is there e.g. a website for software written to make use of BLAS? > d) Are any of the above really ignorant questions? If so please >Enlighten me... > I'll try to take a shot at answering this one (as you say definitions of "best" vary widely). The BLAS libs come in three flavors. BLAS1: Vector operations BLAS2: Matrix-vector operations BLAS3: Matrix-Matrix operations A FAQ on BLAS is available at http://www.netlib.org/blas/faq.html There is a lot of mathematical software that can make excellent use of the BLAS libs -- even a simple operation like scaling a vector by a constant. All of the BLAS implementations for Linux suffer from various shortcomings. 1. The generic fortran routines available at Netlib suffer from lousy performance no matter what compiler you use. 2. Greg Henry's libs are not open source and don't work particularly well on the Athlon. 3. For the Level 3 BLAS the ATLAS package is a good solution under Linux. R. Clint Whaley and Antoine Petitet have released version 3.0BETA of ATLAS which also has some support for generating optimized Level 1 and Level 2 BLAS libs. (You can grab it at http://www.netlib.org/atlas/) The license seems fairly liberal even though it's not GPL. I havn't benchmarked the new release yet but from looking at the source I suspect that the Level1 and Level2 performance is not particularly good yet. The problem is that ATLAS uses GCC which does not take advantage of the new prefetch instructions supported by the PIII and the Athlon. For the Level1 and Level2 BLAS libs the use of these instructions is very important in order to hide main-memory latency. The stuff I wrote was assembly language versions of the Level1 BLAS for the Athlon which use the new prefetch instructions. They're GPL and could probably be adapted for the PIII fairly easily. (I don't have a PIII so I havn't done this). You can get these at http://nemo.physics.ncsu.edu/~briggs/blas_src_v0.11_tar.gz Presumably at some point in the future GCC will be able to generate better code that uses the new instructions and the assembly language won't be necessary but for now thats not the case. For recommendations I would say that ATLAS is a great choice for the Level3 stuff. If you've got an Athlon then my libs are OK for the Level1. No recommendations for Level2 (Our codes don't require much Level2 stuff so it hasn't been a high priority for me). Regards Emil ------------------------------------------------------------------- To unsubscribe send a message body containing "unsubscribe" to beowulf-request@beowulf.org From chemistry-request@server.ccl.net Wed Dec 22 03:45:56 1999 Received: from mail.anrb.ru ([62.76.11.150]) by server.ccl.net (8.8.7/8.8.7) with ESMTP id DAA23171 for ; Wed, 22 Dec 1999 03:45:50 -0500 Received: from mail.anrb.ru (P150 [62.76.11.193]) by mail.anrb.ru with SMTP (Microsoft Exchange Internet Mail Service Version 5.5.1960.3) id ZMDWV559; Wed, 22 Dec 1999 13:35:38 +0500 Date: Wed, 22 Dec 1999 12:43:46 +0500 From: "Ilfir R. Ramazanov" X-Mailer: The Bat! (v1.036) S/N 0 Reply-To: "Ilfir R. Ramazanov" Organization: Institute of Petrochemistry and Catalysis,Ufa,Russia Priority: Normal Message-ID: <13530.991222@anrb.ru> To: chemistry@ccl.net Subject: Re: CCL:Athlon K7 chemistry benchmarks References: <14432.4848.113251.17212@lrz.uni-muenchen.de> Mime-Version: 1.0 Content-Type: text/plain; charset=Windows-1251 Content-Transfer-Encoding: 8bit Hello CCLers! I found really interesting results for g98 (2.3x it's not joke!). Thanks, Eugene. AFAIK Athlon treats non-optimized code as well as Intel optimized (I mean FPU unit). It's a explanation probably. Merry Christmas for everyone! Ilfir ńšåäą, 22 Äåźąįšü 1999, you wrote: EL> Very interesting results for g98 and CHARMm EL> http://www.cpureview.com/art_k7sci1_a.html EL> -= This is automatically added to each message by mailing script =- EL> CHEMISTRY@ccl.net -- To Everybody | CHEMISTRY-REQUEST@ccl.net -- To Admins EL> MAILSERV@ccl.net -- HELP CHEMISTRY or HELP SEARCH EL> CHEMISTRY-SEARCH@ccl.net -- archive search | Gopher: gopher.ccl.net 70 EL> Ftp: ftp.ccl.net | WWW: http://www.ccl.net/chemistry/ | Jan: jkl@ccl.net Best regards, Ilfir R. Ramazanov, Ph.D., Laboratory of Catalytic Synthesis Institute of Petrochemistry and Catalysis, pr. Oktyabrya, 141, Ufa, 450075, Russia. mailto:elf@anrb.ru Visit my homepage and find some QC software http://members.tripod.com/~ChemELF Visit our lab web page http://organomet.cjb.net 22.12.1999 12:35 From chemistry-request@server.ccl.net Wed Dec 22 06:10:20 1999 Received: from bacchus.pc1.uni-duesseldorf.de (root@bacchus.pc1.uni-duesseldorf.de [134.99.152.11]) by server.ccl.net (8.8.7/8.8.7) with ESMTP id GAA23813 for ; Wed, 22 Dec 1999 06:10:19 -0500 Received: from bacchus.pc1.uni-duesseldorf.de (jochen@localhost [127.0.0.1]) by bacchus.pc1.uni-duesseldorf.de (8.8.7/8.8.7) with SMTP id LAA29642; Wed, 22 Dec 1999 11:05:11 +0100 From: Jochen Kuepper Reply-To: jochen@pc1.uni-duesseldorf.de Organization: Heinrich-Heine-Universität To: jon@gate.sinica.edu.tw, Eugene Leitl Subject: Re: Athlon K7 chemistry benchmarks Date: Sat, 28 Jan 2136 17:29:29 +0100 X-Mailer: KMail [version 1.0.28] Content-Type: text/plain Cc: CCL MIME-Version: 1.0 Message-Id: <99122211051100.29612@bacchus.pc1.uni-duesseldorf.de> Content-Transfer-Encoding: 8bit X-MIME-Autoconverted: from quoted-printable to 8bit by server.ccl.net id GAA23814 > Very interesting results for g98 and CHARMm > http://www.cpureview.com/art_k7sci1_a.html What OS (Linux I guess), what compiler, what flags, what libraries ? Is there a comparison to Linux/AXP anybody knows of - esp. using the Compaq fort compiler ? Interesting, anyway :-) -- Jochen Heinrich-Heine-Universität, Institut für Physikalische Chemie I Universitätsstr. 1, Geb. 26.43.02.29, 40225 Düsseldorf, Germany phone 02118113681 fax 02118115195 -- www-public.rz.uni-duesseldorf.de/~jochen Jochen@Uni-Duesseldorf.de -- Jochen.Kuepper@FernUni-Hagen.de -- Kuepper@ACM.org From chemistry-request@server.ccl.net Wed Dec 22 21:02:13 1999 Received: from lrz.uni-muenchen.de (lsajca1-ar3-012-122.biz.dsl.gtei.net [4.3.12.122]) by server.ccl.net (8.8.7/8.8.7) with ESMTP id VAA27897 for ; Wed, 22 Dec 1999 21:02:12 -0500 Received: (from eugene.leitl@localhost) by lrz.uni-muenchen.de (8.8.8/8.8.8) id RAA12007; Wed, 22 Dec 1999 17:50:36 -0800 From: Eugene Leitl MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Message-ID: <14433.32748.683325.273241@lrz.uni-muenchen.de> Date: Wed, 22 Dec 1999 17:50:36 -0800 (PST) To: Rick Venable Cc: Eugene Leitl , chemistry@ccl.net, gfahy@21cm.com Subject: Re: CCL:Athlon K7 chemistry benchmarks In-Reply-To: References: <14432.4848.113251.17212@lrz.uni-muenchen.de> X-Mailer: VM 6.71 under 21.1 (patch 4) "Arches" XEmacs Lucid Rick Venable writes: > And apparently incorrect, at least for G98; the site now has a > retraction. The K7 CHARMM numbers aren't that impressive either-- if I Yes, the author has today contacted me, telling that inadvertely two different g98 versions (optimizations have resulted in a ~2x speedup) have been compared. > scale the P3 result to the 600 Mhz clock speed for the K7, I get an est. > 612 sec, instead of the reported 642 sec. Imo, Athlon has still the following pro points: 1) Price/performance 2) Availability. Typically, commercially available Athlons are ~100 MHz faster than fastest Intel offering available at the same time. 3) Future scalability (Athlon core is designed to go to go whether a Pentium core cannot go, at least without a radical redesign) 4) Current results are likely to get better when compilers support Athlon fully. 5) Investment in a competitor (i.e. not supporting a semimonopoly) will overall result in better products over several silicon iterations Regards, Eugene Leitl From chemistry-request@server.ccl.net Wed Dec 22 21:28:06 1999 Received: from lrz.uni-muenchen.de (lsajca1-ar3-012-122.biz.dsl.gtei.net [4.3.12.122]) by server.ccl.net (8.8.7/8.8.7) with ESMTP id VAA27975 for ; Wed, 22 Dec 1999 21:28:05 -0500 Received: (from eugene.leitl@localhost) by lrz.uni-muenchen.de (8.8.8/8.8.8) id SAA12214; Wed, 22 Dec 1999 18:18:32 -0800 Resent-From: Eugene Leitl Resent-Message-ID: <14433.34424.388261.119069@lrz.uni-muenchen.de> Resent-Date: Wed, 22 Dec 1999 18:18:32 -0800 (PST) Resent-To: , X-Sender: jon@morgan.nmr.ibms.sinica.edu.tw Message-Id: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; CHARSET=US-ASCII Content-ID: From: To: Eugene.Leitl@lrz.uni-muenchen.de Subject: K7 Date: Thu, 23 Dec 1999 09:13:10 +0800 From: Jon Wright [...] Yes pass it on pls, I would rather people know I made a mistake and took the correct results than have the wrong ones. Anyway the things that stand out: On x86 machines with Open shell calcs g98a7 is much faster previous versions K7's are abit faster than Piii's 2x Piii is nearly 1.8x faster than 1 Piii x86 can compete with other machines (I know SGI O200 and O2000 is now mips r12k 300 but the price difference is makes it very unatractive) These results and more can be found at http://www.ibms.sinica.edu.tw/~jon/machines.shtml I have already had a few emails saying that the jobs should be different/bigger but as I said before I started with this input files 3yrs ago when they were consider big then, trying to get hold of all those machines again to run a different job is very hard. The results are enough to show general trends like x86 is comparible to Mips, cyrix/winchip is awful, celerons are reasonable compared to Pii and Piii Regards Jon From chemistry-request@server.ccl.net Thu Dec 23 07:06:01 1999 Received: from orange.csi.cam.ac.uk (exim@orange.csi.cam.ac.uk [131.111.8.77]) by server.ccl.net (8.8.7/8.8.7) with ESMTP id HAA31883 for ; Thu, 23 Dec 1999 07:06:00 -0500 From: ds10025@cam.ac.uk Received: from cugd-pc-152.geog.cam.ac.uk ([131.111.38.143] helo=databug) by orange.csi.cam.ac.uk with smtp (Exim 3.03 #1) id 1215zC-000426-00; Thu, 23 Dec 1999 11:00:46 +0000 X-Sender: ds10025@pop.hermes.cam.ac.uk X-Mailer: Windows Eudora Version 1.4.4 Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" To: , chemistry@ccl.net Subject: mount failed:invalid argument Message-Id: Date: Thu, 23 Dec 1999 11:00:46 +0000 Hi I'm trying to install mandrake from a CD-ROM drive (IDE 48x). But, I'm getting the above error. Any ideas? Dan From chemistry-request@server.ccl.net Thu Dec 23 06:47:06 1999 Received: from hermes.bmc.uu.se (hermes.bmc.uu.se [130.238.39.159]) by server.ccl.net (8.8.7/8.8.7) with ESMTP id GAA31118 for ; Thu, 23 Dec 1999 06:47:05 -0500 Received: from alpha2.bmc.uu.se ([130.238.37.65]) by hermes.bmc.uu.se (Post.Office MTA v3.5.3 release 223 ID# 0-62649U1500L1000S0V35) with SMTP id se; Thu, 23 Dec 1999 11:43:18 +0100 Received: from werkman.bmc.uu.se by alpha2.bmc.uu.se; (5.65v3.2/1.1.8.2/30Jan97-1111AM) id AA15523; Thu, 23 Dec 1999 11:40:38 +0100 Date: Thu, 23 Dec 1999 19:38:36 +0100 (CET) From: David van der Spoel X-Sender: spoel@werkman.bmc.uu.se To: eugene.leitl@lrz.uni-muenchen.de Cc: chemistry@ccl.net, beowulf@beowulf.gsfc.nasa.gov Subject: Re: CCL:K7 In-Reply-To: Message-Id: Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII A comparison of Athlon/PIII to Alpha machines running Linux/Compaq Compilers is also on the gromacs benchmark: conclusion is that a 750 MHz Athlon should be almost as fast as a 466 MHz 21264 Alpha (the cheap ones), however at half the price. But.... if you have parallel code the best buy may still be a Dual Pentium machine. We need SMP Athlon machines... http://md.chem.rug.nl/~gmx/gmxbench.html Groeten, David. ________________________________________________________________________ Dr. David van der Spoel Biomedical center, Dept. of Biochemistry s-mail: Husargatan 3, Box 576, 75123 Uppsala, Sweden e-mail: spoel@xray.bmc.uu.se www: http://zorn.bmc.uu.se/~spoel phone: 46 18 471 4205 fax: 46 18 511 755 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++