olin@CHEME.TN.CORNELL.EDU (Steve Thompson) (05/29/89)
>It's a good thing I know that my numbers are the only right ones :-), >otherwise I might have acted on information like this. > >Jim Helman >Department of Applied Physics P.O. Box 10494 >Stanford University Stanford, CA 94309 >(jim@thrush.stanford.edu) (415) 723-4940 You may (or may not!) be interested in some numbers I have on my application, which is a totally CPU-bound Molecular Dynamics simulation (no I/O, small memory requirements, single precision). I don't have any numbers for the SGI R2000 boxes, but a rough estimate could be obtained from the DECstation 3100 times (DS3100 is R2000/R2010, 16.67MHz). The speeds are all relative to a VAX 8250 = 1.00, and are arranged in descending order of speed. I have some preliminary numbers for the SPARCstation 1 (approx. half of the DECstation 3100); these are not in the table because they are not final. Steve --------------------------------------------------------------------------- Machine Relative Speed Notes --------------------------------------------------------------------------- Cray-1 82.4 fully vectorized IBM 3090-600 35.1 (opt(3) vector FPS 264 15.6 IBM 3090 host DECstation 3100 8.0 f77 -O2 VAX 8700 4.42 VMS V4.7 VAX 8530 3.12 VMS V4.7 Sun 4/260 2.54 f77 -O VAXstation 3200 2.47 VMS V4.7 VAXstation 3100 2.44 VMS V5.1-B Sun 3/160 1.49 OS 3.5 with fpa VAX 8250 1.00 VMS V4.7 MicroVAX II 0.64 VMS V4.7 Sun 3/60 0.44 OS 3.5, no fpa Sun 3/50 0.33 ditto ---------------------------------------------------------------------------
berkley@wucs1.wustl.edu (Berkley Shands) (01/10/91)
I thought some sites on the net would like to see a different benchmark ranking of some current machines... Berkley Shands The following is a floating point benchmark used by our group to rate CPU speed on a standard problem set in molecular modeling. The data set is the Angiotensin Converter Enzyme Inhibitors. The application is an in-house conformational search package written in C. The Rotatable bond scan factor is 4 degrees. Runtime summary of Constrained Search (CSR) V4.2 from The Center For Molecular Modeling, Washington University - St. Louis, Mo. Optimization level as indicated, single precision floating point when available. Micro-VAX-II runs VMS 5.3-2, all others are running some flavor of BSD or SYSV Unix. Single stream execution on multiprocessors. The SPARC machines used an aproximate square-root. All others used the standard system call. This is equivalent to a 20% speedup on SPARC machines. ******************************************************************************* CPU Secs X uVAXII CPU ======== ======== ========================= 289 35.98 33Mhz R3000 SGI 4d/380 -O2 378 27.51 24Mhz R3000 DECstation-5000/200 -O3 410 25.36 24Mhz R3000 ESV -O3 457 22.75 IBM R6000-320 Xcl -O (not fully optimized) 504 20.63 20Mhz R3000 Personal Iris -O3 592 17.56 16Mhz R3000 4d/80s Iris -O3 717 14.50 4/380 Sun Sparc -O4 857 12.13 12Mhz R2000 Personal Iris -O3 1017 10.22 Sun Sparcstation SLC -O4 1054 9.86 4/280 Sun Sparc -O4 2308 4.50 VAXStation 3520 10400 1.00 Micro-VAX-II *******************************************************************************
khb@chiba.Eng.Sun.COM (Keith Bierman fpgroup) (01/10/91)
In article <1991Jan9.191443.4903@cec1.wustl.edu> berkley@wucs1.wustl.edu (Berkley Shands) writes:
The SPARC machines used an aproximate square-root. All others used the standard
system call. This is equivalent to a 20% speedup on SPARC machines.
Could you explain this ? SPARC machines (post the 4/110, 4/2xx) have
hw sqrt.... perhaps you neglected to specify -cg89 -libmil (or
equivalent) in your runs.
--
----------------------------------------------------------------
Keith H. Bierman kbierman@Eng.Sun.COM | khb@chiba.Eng.Sun.COM
SMI 2550 Garcia 12-33 | (415 336 2648)
Mountain View, CA 94043