Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!ames!decwrl!ucbvax!ucsfcgl!evensen
From: evensen@cgl.ucsf.edu (Erik-Robert Evensen%Kollman)
Newsgroups: comp.graphics
Subject: Re: Personal IRIS
Summary: I like the IBM!
Keywords: performance, living with AIX, etc
Message-ID: <14885@cgl.ucsf.EDU>
Date: 4 Aug 90 05:32:33 GMT
References: <850@earth.cs.utexas.edu> <7009@helios.TAMU.EDU> <1990Aug01.170559.623@spatial.com>
Sender: daemon@cgl.ucsf.edu
Reply-To: evensen@cgl.ucsf.edu (Erik-Robert Evensen)
Organization: Computer Graphics Lab, UCSF
Lines: 83

I'm new at posting and I just started reading comp.graphics today.

Now, I admit that I am not a hard core graphics person (despite my
affiliation with CGL) -- my degree is in biochemistry and most of
my background is in writing "numerically intensive" molecular dynamics
simulation code.  However, for the past 9 months I have been writing
an application which serves as a front end and model builder for our
MD code.  This application, by nature, is interactive and I have been
working on the graphics interface part of the program for the past
month and a half.  In the mind of fairness and complete disclosure, I
should say that I worked for IBM last summer and I still work
very closely with some people at the IBM Palo Alto Scientific Center
and count some IBMers among my close friends.  I also feel that I have
been treated very well by the SE's at the San Francisco branch office.
Now, I'm sure that this makes me somewhat biased toward the IBM machine
but I would not put my "professional reputation" on the line my posting
out and out lies about the IBM RS/6000 systems.

That said, I feel compelled to post an opinion which is somewhat
different from what I have seen expressed in this news group regarding
the IBM RS/6000.  I have had a pre-release RIOS (model 320) on my desk for
about 3 months now and I am *very* pleased with the machine.  I've had very
few problems with the machine and my productivity has increased markedly
since I got it.  The only real problem I've had is that the machine
did not recognize anything beyond it's subnet after I rearranged some
of the boards to install the "SGI" cards.  This problem with solved quickly
by the SE who is my contact.  I'm not anything resembling a UNIX wizard
and the SMIT tool has made administering the system quite easy and feasible
for me.  The OS (version 9013) has proven to be fairly stable -- and I
have beat on it pretty hard.

The performance for what I do is almost untouchable when compared to the
other workstations I have seen and used.  For my MD code, it runs about
3-4 times faster than our SS1's and DECstation 3100.  On the same code,
the RIOS runs about the same speed as an IBM 3090 model 300e running
scalar code -- this machine is no slouch in terms of FP performance.
We also have IRISes (IRII?) at our site but I do not use them very
extensively.  I have benchmarked my MD code on the PI and 4D/GT70
and found that the RIOS runs appx 3 times faster than these machines.  All of
the above observations are, of course, problem dependent.  I can't really
compare the interactive graphics performance since I don't have a graphics
application which runs on both systems.  My seat of the pants impression
is that the RIOS is snappier in terms of interactive performance but many
may disagree with me; I've only used the IRIS for a short period of time
and displaying a very small molecule so my comparison is "uninformed" --
take it with a huge grain of salt.

Perhaps the most impressive demonstration of the power of the IBM RS/6000
is in my application (the model builder) I have code which does some of the
numerically intensive calculations from the MD code and displays the
results (building a PHIGS structure for each iteration) in nearly real
time -- that is to say the motion looks almost smooth.  I've had another
user logged into the machine running the numerical intesive code in the
foreground while I was on the console using X windows doing a *large*
compile in one window and running my program using the virtual sphere
controller to do rotations of a molecular wireframe at the same time --
this represents a relatively large load for the system and it performed like
a champ.  Of course there was degradation of interactive response but it
was by no means unacceptable.

The compilers and other development tools are solid in the 9013
release of the code and I expect them to be better in the Golden Code
which I'm supposed to get early next week.  One of the people I work
with (an IBMer) did find a small bug in the xlf compiler in the 9005
release and it was fixed.  I have yet to find a problem with the xlc
compiler and I have about 25 000 lines of not-always-pretty code in
my model builder.  As far as whether AIX looks like UNIX or not, from
my standpoint it is a very familiar environment and I learned most of
my UNIX on a VAX 8650 running BSD and SparcStations running SUNOS.
I'm relatively new to UNIX (came from VM/CMS and VMS) and don't do any
system programming so my views on this matter may not reflect those
of others who are more experienced and / or do "system stuff".

I'm sorry this turned into such a long winded exposition but I hope
it is useful and informative.  I welcome any discussion about the
new IBM machines.  To summarize my impressions in a phrase --
I'm very impressed with the new machines and would buy one (or
several) if it were up to me.

Erik-Robert Evensen
Computer Graphics Lab
University of California San Francisco
coming soon to the ivy covered walls in Cambridge, MA...