de5@ornl.gov (Dave Sill) (12/06/90)
The latest unofficial table of results for the controversial trivial
UNIX benchmark:
echo 2^5000/2^5000 | /bin/time bc
Notes: The time figure reported is "real", not "user"
Report corrections/additions to de5@ornl.gov
Some results are vendor-supplied
Disclaimer: These results should not be used to specify minimum
performance requirements in a procurement specification. This
benchmark is best used as a diagnostic tool, e.g., compare successive
runs on on machine or compare similar machines. It can also be used
to determine the rough capabilities of an unknown system when more
thorough and accurate testing isn't feasible.
SYSTEM TIME REPORTER
------ ---- --------
Fujitsu M780/10 UTS 1.9 toyama@ae.keio.ac.jp
MIPS RC6280 2.8 kmk@tut.fi
IBM RS/6000 Model 530 3.4 mccalpin@perelandra.cms.udel.edu
IBM RS/6000 Model 530 3.5 xxdon@monet.lerc.nasa.gov
IBM RS/6000 Model 730 3.5 tif@doorstop.austin.ibm.com
Alliant FX/2800 3.6 turner@sp64.csrd.uiuc.edu
MIPS RC6280 4.0 mark@mips.COM
Amdahl 5870 (UTS) 4.3 xxdon@monet.lerc.nasa.gov
IBM RS/6000 320 4.4 de5@ornl.gov
Cray Y-MP (6ns) 5.0 elm@sprite.Berkeley.EDU
SGI 4D/380SX 5.2 kmk@tut.fi
SGI 320 (2 cpu) 5.2 xxdon@monet.lerc.nasa.gov
SGI 340 (4 cpu) 5.2 xxdon@monet.lerc.nasa.gov
Motorola 8612 33MHz 88k 5.2 matth@oakhill.sps.mot.com
HP9000 series 870 5.5 dvl@hpcupt1.cup.hp.com
Cray YMP4/64 5.9 xxdon@monet.lerc.nasa.gov
DECstation 5000/200 6.0 de5@ornl.gov
DECstation 5000/200 6.0 NEUMANN@awiwuw11.wu-wien.ac.at
Stardent 3000 6.0 kmk@tut.fi
DECsystem 5810 6.2 de5@ornl.gov
MIPS RC3280 7.0 kmk@tut.fi
DECsystem 5400 (J4.0) 7.6 doi@jrdmax.jrd.dec.com
Cray XMP4/8 7.8 xxdon@monet.lerc.nasa.gov
DG 88k at 25 Mhz 8.1 philip@beeblebrox.dle.dg.com
SGI 4D25 8.1 xxdon@monet.lerc.nasa.gov
Cray XMP (Unicos) 8.6 de5@ornl.gov
Cray 2 (4.1ns) 8.8 elm@sprite.Berkeley.EDU
Sun 4/470 9.1 casper@fwi.uva.nl
Convex C220 9.4 xxdon@monet.lerc.nasa.gov
DECstation 3100 (4.0) 9.7 NEUMANN@awiwuw11.wu-wien.ac.at
DECstation 3100 (3.1) 9.8 NEUMANN@awiwuw11.wu-wien.ac.at
Compaq 486/25 ISC 2.2 11.2 suitti@ima.isc.com
VAX 8800 (v4.0) 12.1 doi@jrdmax.jrd.dec.com
SPARCstation 1+ 12.2 de5@ornl.gov
Sun 4/370 12.3 casper@fwi.uva.nl
Compaq 386/33 ISC 2.2 12.5 suitti@ima.isc.com
SPARCstation 1+ 12.6 casper@fwi.uva.nl
DECstation 2100 12.8 de5@ornl.gov
VAX 6410 (4.0) 12.8 doi@jrdmax.jrd.dec.com
DECstation 2100 13.4 meissner@osf.org
MicroVAX 3900 (J4.0) 13.5 doi@jrdmax.jrd.dec.com
HP9000/375, HP-UX 7.0 14.4 wunder@orac.HP.COM
VAX 6320 (J4.1) 15.1 doi@jrdmax.jrd.dec.com
SPARCstation SLC 15.4 juan@burdell.gatech.edu
SPARCstation SLC 15.4 casper@fwi.uva.nl
SPARCstation 1 16.0 juan@burdell.gatech.edu
Compaq 386/25 ISC 2.2 16.5 suitti@ima.isc.com
HP9000-835CHXSE (7.0) 18.2 jsadler@misty.boeing.com
HP9000/835 (HP-UX 7.0) 20.6 NEUMANN@awiwuw11.wu-wien.ac.at
Sun 3/60 31.0 casper@fwi.uva.nl
SGI 3030 37.8 xxdon@monet.lerc.nasa.gov
Sun 3/80 38.3 casper@fwi.uva.nl
SPARCstation SLC 44.5 karplus@ararat.ucsc.edu
INTeL 80286 SCO 45.8 ethan@thinc.UUCP
Sun 3/50 45.9 zeeff@b-tech.ann-arbor.mi.us
IBM RT 48.8 zeeff@b-tech.ann-arbor.mi.us
MicroVAX 2000 (J4.1) 74.1 doi@jrdmax.jrd.dec.com
MicroVAX 2000 75.7 de5@ornl.gov
Zenith 386/20 ISC 2.2 86.6 suitti@ima.isc.com
--
Dave Sill (de5@ornl.gov)
Martin Marietta Energy Systems
Workstation Supportwunder@orac.HP.COM (Walter Underwood) (12/07/90)
Dave Sill sez: Notes: The time figure reported is "real", not "user" ... Excuse me, but this is stupid. Some of these machines were loaded, and some were not, so please use "user time". Wall time (real) is only meaningful if all of the load on the system is part of the benchmark. That is not true for this benchmark, so your list is invalid. If you insist on using real time, please delete the entry that I reported. wunder
ken@ab.msc.umn.edu (Ken Chin-Purcell) (12/07/90)
de5@ornl.gov (Dave Sill) writes:
The latest unofficial table of results for the controversial trivial
UNIX benchmark:
echo 2^5000/2^5000 | /bin/time bc
An IBM 3090 600 Model J, running AIX MP:
sg> echo 2^5000/2^5000 | /bin/time bc
1
real 1.9
user 1.7
sys 0.0
World's fastest 'bc' machine?
--
-- Kenlih@cbnewsk.att.com (andrew.a.lih) (12/07/90)
In article <3083@uc.msc.umn.edu>, ken@ab.msc.umn.edu (Ken Chin-Purcell) writes: > An IBM 3090 600 Model J, running AIX MP: > > sg> echo 2^5000/2^5000 | /bin/time bc > 1 > > real 1.9 > user 1.7 > sys 0.0 > > World's fastest 'bc' machine? Not quite, my AT&T 6386/25Mhz running Coherent wallops it: Real: 0.3 User: 0.1 Sys: 0.1 Anyone care to explain this? My guess is that Mark Williams Co. who puts out Coherent really tweaked something! Results for 2.1^5000/2^5000 were quick as well: Real: 1.3 User: 1.1 Sys: 0.1 The same hardware running AT&T SystemV/386 v3.2.2 got: Real: 16.6 User: 16.3 Sys: 0.0 Andrew Lih AT&T Bell Laboratories (lih@probe.att.com)
john@hopf.math.nwu.edu (John Franks) (12/07/90)
In article <1990Dec6.230423.13969@cbnewsk.att.com> lih@cbnewsk.att.com (andrew.a.lih) writes: >In article <3083@uc.msc.umn.edu>, ken@ab.msc.umn.edu (Ken Chin-Purcell) writes: >> An IBM 3090 600 Model J, running AIX MP: >> >> World's fastest 'bc' machine? > >Not quite, my AT&T 6386/25Mhz running Coherent wallops it: > > Real: 0.3 > User: 0.1 > Sys: 0.1 > >Anyone care to explain this? My guess is that Mark Williams Co. >who puts out Coherent really tweaked something! Results for >2.1^5000/2^5000 were quick as well: [Coherent bc over 10 times as fast as AT&T version on same hardware] > >Andrew Lih >AT&T Bell Laboratories >(lih@probe.att.com) Coherent is a Version 7 clone reverse engineered from scratch almost ten years ago (anyway I seem to recall it predates IBM PCs -- it ran originally on PDP 11s). Internally the bc program bears no relation to the real UNIX version. The Coherent version was written by Henry Cejtin who went on to become one of the authors of Mathematica (he recently has left Wolfram Inc.) The explanation for the benchmark result above is simple. Cejtin's bc is much better than the original or whatever version AT&T is using now! This is a fairly dramatic demonstration that various programs which happen to share the name "bc" and the functionality of a calculator don't necessarily make good benchmarks. John Franks Dept of Math. Northwestern University john@math.nwu.edu
de5@ornl.gov (Dave Sill) (12/10/90)
In article <19040002@orac.HP.COM>, wunder@orac.HP.COM (Walter Underwood) writes: > > Dave Sill sez: > Notes: The time figure reported is "real", not "user" ... > >Excuse me, but this is stupid. You're not excused. There's no need to be rude about it. If you have something to say, then say it. If you disagree, present the facts and let them make your point. >Some of these machines were loaded, >and some were not, so please use "user time". Wall time (real) is >only meaningful if all of the load on the system is part of the >benchmark. That is not true for this benchmark, so your list is >invalid. As Michael Meissner pointed out in his article under the "Re: bc is bs" subject, it sometimes makes more sense to benchmark systems in their usual state. I agree that it's also useful to have some idea of how the system load has affected the results, so I've updated my table to include real, user, and system time, where available. -- Dave Sill (de5@ornl.gov) Martin Marietta Energy Systems Workstation Support