From chemistry-request@server.ccl.net Mon Apr 26 11:04:36 1999 Received: from relay.ppco.com (relay.ppco.com [207.27.254.3]) by server.ccl.net (8.8.7/8.8.7) with SMTP id LAA02730 for ; Mon, 26 Apr 1999 11:04:32 -0400 Received: by relay.ppco.com id JAA02481 (InterLock SMTP Gateway 4.2 for chemistry@ccl.net); Mon, 26 Apr 1999 09:59:56 -0500 Message-Id: <199904261459.JAA02481@relay.ppco.com> Received: by relay.ppco.com (Internal Mail Agent-1); Mon, 26 Apr 1999 09:59:56 -0500 Date: Mon, 26 Apr 1999 09:59:54 -0500 From: "George D. Parks" X-Mailer: Mozilla 4.51 [en] (WinNT; U) X-Accept-Language: en Mime-Version: 1.0 To: chemistry@ccl.net Subject: GAMESS Timings on SGI NT Workstation Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit With all the recent discussions of the pros and cons of benchmarking, I'm hesitant to send these results out, but last week Steven Creve asked about timings on the new SGI NT workstations. We had access to a loaner machine for a brief time and ran some GAMESS timings. Here are the results from several computers. Timing for the benchmarks is given below the table, but here are a few general comments. 1. PCGAMESS (NT) is significantly faster than the Linux version. The NT version of GAMESS has been fine tuned by Alex Granovsky, so one can't really assign the the difference to any single factor without more work. 2. The standard office IBM PC looks good for cpu, but the I/O system really takes a hit when there's much disk I/O. 3. The SGI PC is fast! CPU utilization was best of any machine tested. This is especially obvious on the thymine job which created a 665MByte integral file (much of Granovsky's tuning was focused on better integral storage). 4. The SGI R10K machine wasn't completely unloaded. It's a dual processor machine, and G98 was running on the other processor while I was running these jobs, so that may have affected some of the times. George Parks Phillips Petroleum Company All opinions expressed here are mine and not those of Phillips Petroleum Company. =========================== CPU Time (seconds) Thymine Thymine C4H6 Ti2H8 C4H4 Hexene (conv) (dir) IBM PC 377 1,005 172 337 267 138 NT PC 531 1,360 234 469 358 183 LinuxPC 801 1,420 427 753 593 503 R10K 476 1,160 301 588 457 279 SGI PC 366 912 248 324 163 127 Wall Clock (seconds) Thymine Thymine C4H6 Ti2H8 C4H4 Hexene (conv) (dir) IBM PC 2,541 1,010 180 805 298 206 NT PC 1,207 1,362 235 476 359 183 LinuxPC 1,690 1,422 429 1,493 594 1,413 R10K 1,504 1,190 311 1,483 479 283 SGI PC 393 913 249 337 163 127 CPU Utilization Thymine Thymine C4H6 Ti2H8 C4H4 Hexene (conv) (dir) IBM PC 14.8% 99.5% 95.6% 41.9% 89.4% 66.7% NT PC 44.0% 99.9% 99.6% 98.6% 99.7% 99.8% Linux PC 47.4% 99.9% 99.6% 50.4% 99.8% 35.6% R10K 31.6% 97.5% 96.8% 39.6% 95.5% 98.7% SGI PC 93.1% 99.8% 99.3% 96.2% 99.5% 99.4% =========================== I used some GAMESS input files that have been used for benchmarking by Shiro Koseki in Japan. I also added an MP2 energy calculation for 1-hexene. Input files are available. =========================== PC Gamess Home Page: http://classic.chem.msu.su/gran/gamess/index.html GAMESS Home Page: http://www.msg.ameslab.gov/gamess/GAMESS.html GAMESS Benchmarks: http://xenon54.edu.mie-u.ac.jp/~shiro/gms-bench.html =========================== Computers (1 cpu used on all machines) IBM PC 400 MHz IBM 300 PL, 192 MBytes memory, 6.4 GByte IDE disk drive (NTFS), NT 4.0 SP4 NT PC 300 MHz Dual Processor Dell Optiplex, 256 MBytes memory 2x4GByte SCSI disk drives, NT 4.0 SP4 (hardware on this and Linux PC are identical) Linux PC 300 MHz Dual Processor Dell Optiplex, 256 MBytes memory 2x4GByte SCSI disk drives, running Linux kernel 2.2.3, GAMESS compiled with Portland Group FORTRAN compiler R10K Dual 195 MHz processors, 512 MBytes memory, 9 MByte SCSI disk drive SGI PC Model 320-Dual 450 MHz PII processors, 1GByte memory, SCSI disk drive, NT 4.0 SP4 (demo loaner machine from SGI)