Subject: User Impression of SPARCstation 1
From the Sun User Group:
--------------------------------------------------------------------------------


	  The SPARCstation 1: A User's First Impressions

                      Steven M. Christensen
                    Director, Sun User Group

                   Senior Research Scientist,
         National Center for Supercomputing Applications
      Beckman Institute for Advanced Science and Technology
           University of Illinois at Urbana-Champaign
                       steve@ncsa.uiuc.edu

I remember a long while ago tearing open an Apple II box that had just
arrived in a colleague's office.  It was the first microcomputer either
of us had had a chance to play with.  The machine was small and
relatively light weight.  We decided the first thing to do was open it
up and stare at the chips. Inside was a small circuit board, a power
supply and some slots for otherboards.   It amazed us that this little
machine would free us from the dreaded mainframe in the basement of our
building.

Unpacking and Cabling

This scene repeated itself a few weeks ago.  I got a call from our
shipping room here at NCSA telling me that five boxes from Sun
Microsystems had arrived.  I drove my car to the loading dock and
prepared myself for a morning of unloading heavy boxes and digging
through mountains of documentation.  When I got to the shipping room,
the Sun boxes were neatly stacked next to some boxes containing a Mac
IIx and its parts.  Surprisingly, the Sun stack was smaller.  I went to
pick up one box and it was possible to carry it in one hand.  I looked
at the packing slip on the box and it indicated that this was the CPU
unit.  I had seen the SPARCstation before - in Andy Bechtolsheim's lab
long ago and at the San Francisco Civic Center introduction - but had
never picked one up.  The gray-scale monitor box was a bit heavier, but
I could also carry it using one hand.  I, not being a customer of Body
by Jake, Inc., was relieved and grateful that I would not be required
to lug a ton of iron and silicon around for the next few hours.

   Once at my office, I took the components out of their boxes.  At
first I thought that the documentation was missing. Then I found three
small spiral-bound books: SPARCstation 1 Installation Guide, Sun System
& Network Manager's Guide, and Sun System User's Guide.  I immediately
put them aside.  Enjoying adventure, I never use manuals to install a
machine. The books looked interesting - especially since I didn't have
to go out to find some 3-ring binders - and I told myself I would get
to them later. I like reading manuals; everyone has their
peculiarities.

The CPU was the first out of its box. I checked out the "pizza box"
claim. In fact, Andy Bechtolsheim's size analogy is pretty close
(depending on your appetite).  I picked it up and walked down the hall,
holding the unit on my fingertips waiter-style, with the express
purpose of making my colleagues, who are still pushing around 3/160's
and Mac II's, envious.

Returning from my gloating trip, I placed the CPU unit on my desk and
got out the keyboard (the new Sun style with a slightly mushier feel to
it and an unfortunate placement and size of certain control keys),
mouse, monitor and the tape drive unit.  It took me about five minutes
to understand the new cabling system.  The keyboard and mouse use the
Mac-like plugs and are trivial to install.  In fact, all the cables
were smaller and seemed to connect with a lot less hassle.  In other
words, I didn't have to look for a screwdriver.

Looking Inside

Finally, before I plugged it in, I couldn't resist opening the back of
the CPU unit and looking at the 8.5-inch by 11-inch motherboard.  My
first reaction was to compare it to that Apple II and to the first IBM
PC I took apart.  Here I was with a 12.5-MIP machine, 8 megabytes of
memory, 208 megabytes of disk, a color frame buffer and a lot of other
serious stuff, and it looked 10-times as simple and clean as those old
machines.  My next thought was, "Boy, will this make a great lap-top
machine someday."

Another reason for opening the CPU box was that my machine was sent
with two SCSI disk drives and a floppy disk drive that required
installation.  It was so obvious how to install them that, again, I
didn't have to read any instructions except when I needed to make sure
the device numbers were set up correctly and that the terminator plug
on the external tape box was on properly.  The ease of installation
contrasts with other disk installs I have done which require ripping
the machine apart to get at the connectors and screws.  I have heard
that Sun simplified the design of this package to facilitate the use of
robots for manufacturing. This makes it easy for fumble-fingered
carbon-based life to put it together also.

The SPARCstation 3.5 inch floppy drive is MS-DOS compatible and hence
gives software developers and users a cheap and compact way to ship
four megabytes of files.  Since I am developing a commercial package
based on Mathematica (TM), this inexpensive alternative to 1/4-inch
tapes and the infamous optical drives of another vendor makes the
SPARCstation 1 very attractive as a development machine.

The motherboard's most interesting feature is the small number of
chips, the most prominent of which are the SPARC CPU and Weitek
floating-point chips.  This floating-point processor is different from
the one in the high-end SPARCstations.  The Texas Instruments chip used
in those machines is a newer and more costly, but is not significantly
faster than the Weitek chip at the SPARCstation system performance
levels.

Three S-bus slots are available.  For my use, these will be taken up by
a framebuffer connected to a gray-scale monitor, and the GX graphics
boards connected to a 16-inch color monitor.  I will comment on the
performance of these boards later.  Lastly, there are two sets of slots
for memory which consists of 1-megabit SIMMS.  These are easy to access
and install compared with a Mac II, for instance.

Software Installation and Testing

Once all the hardware was but back together and cables connected, the
power was turned on.  The 4.0.3 operating system comes on two 1/4-inch
tapes or on 23 floppies.  The installation was very straightforward and
took about 30 minutes including tape read time.  I took another 30
minutes or so to add users and mount NFS filesystems to the Sun-4/260
machine I have been using for research.  I had no trouble accessing the
software on these mounted file systems.  In particular, I installed
version 1.2 of Mathematica, the software that I use the most.

I now set out to run some benchmarks on the six workstations I use in
my office.  These benchmarks are ones collected either from my own
research in quantum gravity theory or simple Mathematica commands that
test the various kinds of computing.  The following table shows the raw
number results for the benchmarks I ran. Units are seconds, with the
ration of results to a Mac II given in the parenthesis under each
actual time result. The ratio to a Sun-3/160 is given for the tests
that could not be run on a Mac II.

Table.  Benchmark Results - Units in Seconds (Ratio of result compared
	to Mac II in parentheses)

Machines:  Mac II   Sun-3/160   NeXT    Sun-4/110  Sun-4/260 SPARCstation 1 

Benchmarks:
Integer    6.45        4.03     3.33      3.75       2.77         2.30
	  (1.00)      (1.60)   (1.94)    (1.72)     (2.33)       (2.80)

Float     14.58        8.05     5.13      1.88       1.35         1.17
	  (1.00)      (1.81)   (2.84)    (7.76)    (10.80)      (12.46)

Factor    77.58        2.15    28.47      0.83       0.58         0.57
	  (1.00)     (36.08)   (2.72)   (93.47)   (133.76)     (136.11)

Graphics  41.85       20.55    14.92      7.43       5.07         4.50
	  (1.00)      (2.04)   (2.80)    (5.63)     (8.25)       (9.30)

Symbol 1  Bombed    1439.93     N/A     485.40     340.05       342.83
		      (1.00)             (2.97)     (4.23)       (4.20)

Symbol 2   N/A       382.70     N/A     137.10      90.10        92.20
		      (1.00)             (2.79)     (4.25)       (4.15)

Configurations:

Mac II:     	5-Mbytes RAM, color monitor, 80-Mbytes disk, Mathematica 1.1

Sun-3/160:  	16-Mbytes RAM, color monitor, 71-Mbytes disk,
		Mathematica 1.2 beta

NeXT:       	8-Mbytes RAM, grayscale monitor, 330-Mbytes disk,
		Mathematica 1.1

Sun-4/110:  8-Mbytes RAM, high-resolution monochrome monitor, 468-Mbytes disk,
	    Mathematica 1.2 beta

Sun-4/260:  64-Mbytes RAM, high-resolution monochrome monitor and colors
	    monitor, 881-Mbytes disk, Mathematica 1.2 beta
SPARCstation 1GX: 8-Mbytes RAM, grayscale monitor, 208-Mbytes
	    disk, Mathematica 1.2 beta


The Integer benchmark is the Mathematica command:

   1989^1989;//Timing

The Float benchmark is the Mathematica command:

   Eigenvalues[Table[Random[],{30},{30}]]//Timing

The Factor benchmark is the Mathematica command:

   Factor[Expand[(x+y)^30]]//Timing

The Graphics benchmark consists of the Mathematica commands:

   <<Graphics/Polyhedra.m
   Show[Graphics3D[Stellate[Stellate[Icosahedron[]]]]]//Timing

The Symbol 1 is a very large Mathematica calculation which does a set
of tensor manipulations to compute an important object using the curved
space renormalization theory. The Symbol 2 calculation is another
curved space calculation which computes, using a set of C programs, a
very large equation with hundreds of terms. Neither of these two
benchmarks do any floating-point calculations.

The Symbol 1 calculation bombed due to too little memory on the Mac
II.  I had to restart the system to get back to Mathematica.  Since I
had just upgraded the NeXT machine to the 0.9 level of the NeXT
operating system and could not get the networking to work, I could not
run either Symbol benchmark.  My Mac II does not have a C compiler so I
could not run the Symbol 2 program.

I also ran the usual Dhrystone (2.0) and Linpack benchmarks, using
SunFORTRAN 1.2  with a highly optimized set of f77 flags on all the
SPARCsystems.  The SPARCstation 1 ran at 20,000-Dhrystones/sec and did
2.53-megaFLOPS (single precision) and 1.43-megaFLOPS (double
precision).  I have also run Dhrystone on the SPARCstation 330 with
results of 25,000-Dhrystones/sec. The feeling of the SPARCstation 1 is
very fast. When working with the window system, windows pop up
rapidly.

GX Graphics Evaluation

The next step was to install the GX (code name "Lego") graphics
accelerator board.  Before doing this, I tried a few windowing and
graphics operations so that I could compare the feel with and without
the GX board installed.  I took the file containing this article and
put it into a SunView text editor window.  I then tried scrolling it up
and down using the arrow on the scroll bar of the window.  I also
brought up two complicated Mathematica 3-D graphics and got a sense of
the speed at which the shaded objects were completed.

The installation of the GX board required that I remove the color frame
buffer board and replace this one-slot S-bus board with the two-slot GX
board.  This was a one-minute job.  Making sure that the cgsix0 device
was in the /dev file (using MAKEDEV cgsix while being root in /dev) and
making sure the cgsix device was in my kernel configuration file, I
rebooted the machine.  The difference was immediately noticeable.  A
three dimensional grayscale Sun logo came up next to the boot test
lines indicating that the GX board was working.  The rate at which the
lines were painted across the screen was clearly faster.

Once booted and into the SunView window system, I started bringing up
windows.  This operation is already very fast without the GX board, but
was even faster with it.  Bringing up the test file in the text editor
window, I did a scroll check again.  The scrolling was very fast and
smooth with no jumping, and appeared to be about two-times faster than
without the GX board.  The Mathematica graphics took the same time to
compute as before, but  painted about two-times faster.  One final
benefit I noticed was that the look of the graphics was nicer with the
GX board installed.

This inexpensive, easy-to-install board is an impressive addition to
the SPARCstation 1.  I am not a graphics expert (one of the few people
at NCSA who isn't), but it is clear to me that the GX board with its
unique ability to accelerate both graphics and other screen display
functions is a welcome addition to the Sun line of products.

The NCSA User Base

My SPARCstation 1 is on a network at NCSA. The network includes about
70 other Sun machines of all kinds and perhaps another 200 microsystems
and workstations, and a Cray-X/MP and Cray-2.   Most of our machines
have disks to provide local booting and private program storage and are
connected to a number of Sun-4 servers with many gigabytes of disk.
This network extends over a number of buildings linked via optical
fiber and gateways.

At NCSA, the initial purchasers of SPARCsystems are likely be the
in-house scientific staff who require high-performance compute engines.
One of our hardware philosophies has been to closely integrate
supercomputers to workstations.  In effect, we like to think of the Sun
machines as having Cray floating-point coprocessors.  In my work, which
requires no floating point computations of significance, it is the
integer performance of the SPARCstations that interests me.  I do
calculations using Mathematica that take hours of execution time on a
Sun-4, which is actually better than the performance achieved on the
Cray-2 for integer computations. The Cray-2 performance may change when
the C compiler on the Cray is coupled to the architecture of the
supercomputer in the same way that FORTRAN is today. For now, it is
amusing to upset people with selected benchmarks that run faster on a
desktop machine than on a multi-million dollar machine.  In order to
calm them down, I then show them some vectorized FORTRAN codes that run
50-times faster on the Cray than on a workstation.

Summary

I have found the SPARCstation 1 to be an easy-to-install and use
machine which is quiet and produces little heat.  It is very fast and,
except for cases where I need large amounts of memory or SMD disk
speed, I will be moving to it and away from the Sun-4/260 for my daily
research work.  The price with our university discount is impossible to
beat.  Other departments on our campus have chosen the SPARCstation
over the DECstation 3100 despite large discounts by DEC.

I hope that Sun's robots are well-oiled and geared up to work around
the clock. Sun is going to sell a lot of SPARCstations.