Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!ucsfcgl!seibel
From: seibel@cgl.ucsf.edu (George Seibel)
Newsgroups: comp.sys.next
Subject: Re: kiloflops (lots of kiloflops)
Message-ID: <11590@cgl.ucsf.EDU>
Date: 11 May 89 08:56:07 GMT
References: <10598@cit-vax.Caltech.Edu>
Sender: daemon@cgl.ucsf.edu
Reply-To: seibel@hegel.mmwb.ucsf.edu (George Seibel)
Organization: Computer Graphics Lab, UCSF
Lines: 36

In article <10598@cit-vax.Caltech.Edu> christ@tybalt.caltech.edu.UUCP (Christian L. Keppenne) writes:
]it was tough for me to decide whether to buy a NeXT or the new sparcstation 1
]from Sun. now that i have placed an order for the former, i would like to
]know whether it might ever be able to perform honorably when compared tto
]the latter. the guys from Sun claim that their new child can compute at
]1.5 mflops. has anyone got the figures for the NeXT machine? or, if not,
]for another 25 mhz 68030/68882 machine?
]it wouls be nice if we could count on some floating point accelarators
]to be marketted in the near future. maybe there is already something available
] or under development. has anyone heard of anything concerning this?

  I ran an honest-to-god hard core fortran number cruncher on a NeXT box.
This was a macromolecular mechanics program.  I compiled it on a Sun using
the standard f77 compiler, -O and -f68881.  All I had to do was copy the
binary to the cube and it worked perfectly.  (I've been told this may not
work in later revs of the O/S.)   It ran about like a VMS VAX 780 with fpa.
Not too shabby, but it isn't spectacular for 1989.  Various caveats apply:
1) NeXT has a 68882 flt pt chip;  there are instructions available on it
   that f77 -f68881 would not generate.  I've heard there may be a factor
   of two there, but take this with a big grain of salt.  These kinds of
   things are notoriously data dependent and I'm only repeating hearsay.
2) I can't say how good the optimiser on the sun f77 compiler is - my
   suspicion is that you could do better than that, in fact sun is said
   to have an optional fortran compiler that is better.  It remains to be
   seen what quality of fortran will be available for the NeXT, or maybe
   you do your floating pt. in C... same deal though - how good is compiler?
3) Finally, there are lies, damn lies, and benchmarks...  your mileage may
   vary.  My application has lots of indirect addressing, no transcendentals.

Sun is only claiming 1.5 Mflops for the Sparcstation?  hmmm... well, if they
say so.  I thought it was supposed to be faster than that.  Sounds like
it and the cube could be in the same ballpark.   The NeXT box is not a
personal Cray, but it certainly has respectable floating point and has to
be one of the coolest overall packages on the planet.

George Seibel, UCSF