Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!cs.utexas.edu!wuarchive!zaphod.mps.ohio-state.edu!lavaca.uh.edu!karazm!jet
From: jet@karazm.math.uh.edu (J. Eric Townsend)
Newsgroups: comp.sys.ibm.pc.rt
Subject: I've played with one (was Re: Risc System/6000
Message-ID: <1990Feb21.073230.11534@lavaca.uh.edu>
Date: 21 Feb 90 07:32:30 GMT
References: <90051.101319ANKGC@CUNYVM.BITNET> <24325@princeton.Princeton.EDU> <20635@netnews.upenn.edu>
Sender: nntppost@lavaca.uh.edu (NNTP Posting Service)
Distribution: na
Organization: University of Houston -- Department of Mathematics
Lines: 40

In article <20635@netnews.upenn.edu> bradley@grip.cis.upenn.edu (John Bradley) writes:
>Yes, but has anybody actually PLAYED with these things?  I attended the 
>Philadelphia unveiling and got to try out a 320, a 530, and an X terminal.

(This was on the smallest deskside unit, with 40Mb of ram.)

First off, at the Houston show, nobody was allowed to *touch* the
things, much less get at the keyboard.

Monday, we (some folks at UH) went over to IBM to actually get a personal
demo.  They were rather upset that we had a 1/4" tape with source, ready
to try out.

emacs wouldn't compile. :-) The C compiler was barfing on stray interrupts
and never got through a single .c file, so it had little to do with
bad karma from compiling RMS code on an IBM. :-P

(We couldn't run the X terminal and the console in X at the same time,
because some IBM person didn't allocate enough pty's... :-)

The fortran compiler didn't have the normal unix f77 libs -- all we
wanted was dtime() so that our benchmarking programs would work.

After some quick mucking around on a vt100 clone (the big monitors *still*
aren't in whatever AIX uses for a termcap, so vi is useless from a non-X
or non-dumb terminal environ), we hacked our .f code to not call dtime.
Instead, we used /bin/time and did single runs.

We saw between 5 and 5.2 MFLOPS doing fluid dynamics code (gaussian
elimination, matrix multiplies, big data sets, etc); and about 1/4 - 1/5
the speed of an ETA-10P on some code written for the Cray and ETA.
(Sorry, we don't know what the MFLOPS on our ETA code is, the
ETA is our benchmark for that code. :-)

Oh well.  Maybe the code IBM ships will be in better condition.
--
J. Eric Townsend
University of Houston Dept. of Mathematics (713) 749-2120
jet@karazm.math.uh.edu
Skate UNIX(tm).