Message-ID: <bnews.megatest.186>
Newsgroups: net.micro
Path: utzoo!decvax!decwrl!sun!megatest!dre
X-Path: utzoo!decvax!decwrl!sun!megatest!dre
From: megatest!dre
Date: Wed Feb 16 01:04:50 1983
Subject: Benchmarks of machines at Unicom
Posted: Sat Feb 12 19:34:48 1983
Received: Wed Feb 16 01:04:50 1983


At UNICOM Yin Shih and I took the opportunity to run a simple benchmark  on
most  of the machines on display at the vendor exhibit.  This is an attempt
at organizing the results.

In any benchmark it is important to know 1) what the benchmark tests and 2)
whether  you  care  about  what  the benchmark tests.  An example of what I
would consider to be a poor benchmark for most purposes is  the  Whetstone.
The  Whetstone bashes away at transcendental functions, exercising floating
point load, add, and multiply instructions.  Unfortunately  it  uses  these
instructions  with  a  frequency seldom encountered even in "heavy floating
point grinders" such as SPICE.  So, to make a long story short, we're going
to  tell  you  what  we did and what numbers we got, but we're not going to
make any claims about which machine is the  "fastest"  or  "best  buy."  We
aren't  going  to  try  to  interpret  the results.  Just the facts, ma'am.
Also, we did this on our own and not  as  representatives  of  Megatest--so
flame  at  us  but  don't get mad at the company (the only company ENTIRELY
composed of Dead Heads!  Long Live net.gdead!).

Now that we've covered our a**es, let's get to the fun part.


The benchmark (for better or worse) is:


		main()
		{
			long i;
			for (i=0;i<1000000;i++);
			printf("Done.\n");
		}

This is obviously completely cpu bound and tests 32 bit integer  arithmetic
(or at least increment and compare).

We ran two tests.  These were:

			/bin/time cc tst.c
and
			/bin/time a.out

The compile was quite i/o bound on most systems.  Some machines  which  did
very  well  in  the  loop  had  terrible  disk  access  times.  Others  had
reasonable i/o but slow processors.

The system configurations and prices stated below are, in  most  cases,  as
tested.  In  some cases the prices quoted us by the salesmen were for other
configurations and that is given instead.  In a few cases  the  information
was  unavailable.  These  figures  may  be  wrong,  but  we  tried to be as
accurate as we could be. "Total" is the sum of user and supervisor time for
both  tests.  Significant  digits are what "time" gave us except for "real"
times where I seem to have tacked a ".0" on the end of  some  numbers.  The
numbers were corrected for the Xenix ports on the TRS model 16, Apple LISA,
and Parallel 68000 systems after we had a discussion  with  some  Microsoft
folks.  It  seems  that  the TRS model 16 has a 30 Hz clock and "time" gave
numbers that were off by a factor of two for  these  machines  because  all
three  of  them  are  object  file  compatible  and  they  just transported
utilities over without  recompilation.  The  Microsoft  people  were  quite
knowledgeable and very helpful.  Thanks, guys.

Some systems were running multiuser, so the  "real"  numbers  shouldn't  be
used for comparisons.

The numbers, in increasing "total" order:

					      cc            a.out
					--------------  --------------
System & OS     Price   Config          real user supv  real user supv  Total
-----------     -----   ------          ---- ---- ----  ---- ---- ----  -----
Amdahl 470/V8     na      na            5.49 0.26 0.18  1.29 0.24 0.01   0.69
UTS and VM

SEL 32/87       ~$250K    na            10.0  1.0  2.5   1.0  0.8  0.0   4.3
32V

VAX 11/780        na    Emulex SC21     28.0  0.6  1.8  13.0  3.9  0.2   6.5
w/o FP accel.           disk controller
4.1BSD                  w/ CDC 9766s
(courtesy of            1-2MB memory
Rick Kiessig at         (Rick wasn't sure)
Fortune Systems)

Masscomp        $28000  dual 10 MHz      5.0  1.2  2.1   8.0  7.5  0.1  10.9
SIII +                  68000s (for
Berkeley                VM) with a
extensions              4KB cache
			27 MB winch
			1 MB floppy
			512KB memory
			ascii terminal

VAX 11/750     ~$100K   SC21 w/4 160MB  10.0  0.9  2.4   9.0  8.0  0.3  11.6
w/o FP accel            Fujitsu winch
4.1aBSD                 2 MB mem
			Emulex DH
			(the system I'm
			using now)

Perkin-Elmer 3210 ~$50K 64 MB winchester 6.0  1.0  2.8   9.0  8.7  0.2  12.7
V7+Berkeley extensions  512KB memory
+SCCS

Parallel 68000  $14K    10MHz 68000     20.0  2.2  3.4   9.0  7.6  0.0  13.2
Xenix                   8 serial ports
			Multibus
			16MB winch
			1 MB floppy
			256KB memory

HP-9000        $88000   10MB winch      26.8  2.4  6.8   4.4  4.2  0.1  13.5
HP-UX (SIII)            64MB winch
			2 MB memory
			512KB floppy
			plotter
			color graphics

Sun            $29,500  Sun Terminal    14.6  1.4  4.5   9.4  7.8  0.6  14.3
4.1cBSD                 1 MB memory
			84MB winch
			DCD tape drive
			10 MHz 68000
			mouse and
			upgrade to 68010
			(The salesman said
			this would cost $2.5K)

Charles River  $19,400  12.5MHz 68000   26.7  6.1  3.4   6.3  5.7  0.2  15.4
UNOS                    4KB cache
			1MB memory
			4 serial ports
			10 MB winch
			1MB floppy
			Versabus

Alcyon APX    $29000    512KB memory    17.0  2.7  1.0  14.0 13.1  0.0  16.8
Regulus                 winchester
			(size n/a)
			68000
			(speed n/a)
			streamer tape
			drive
			Q bus

Codata 3300    $14000   68000           20.0  3.6  4.5  11.0 10.5  0.1  18.7
V7                      (speed n/a)
			2 serial ports
			12 MB winch
			1 MB floppy
			320K memory

Pacific Micro  $13,900  68000           30.0  3.5  4.6  11.0 10.5  0.1  18.7
Super Sun               (speed n/a)
Unisoft V7              20 MB winch
			1MB floppy
			Multibus
			256KB memory

Onyx C8002     $17,000  4 MHz Z8000     13.1  2.4  3.9  13.5 13.2  0.1  19.6
V7+Berkeley             20 MB winch
extensions              DCD tape drive
			256KB memory

VAX 11/730       na     2-RL02 drives   14.0  2.1  4.1  33.0 14.3  1.0  21.5
4.1BSD                  memory size n/a

Alcyon APS     $9950    512KB memory    19.0 10.1  0.3  12.0 11.4  0.0  21.8
Regulus                 3 serial ports
			10MB removable
			winchester
			8 MHz 68000+
			68451 MMU
			1 wait state

Zilog S8000   $17000    Z8001           28.0  2.2  4.7  16.0 14.9  0.1  21.9
Model 11                (speed n/a)
Zeus (V7 +              18MB winchester
Berkeley                DCD tape
extensions)             ZBI bus
			256K memory

Altos 586      $7990    10 MHz 8086     18.0  1.0  4.3  17.0 16.7  0.2  22.2
Xenix                   512K memory
			minifloppy
			10 MB winch
			5-8 users

Wicat 68000   $17000    8 MHz 68000     26.0  2.8  5.8  15.0 14.7  0.1  23.4
Unisoft                 up to 3 users
V7+SIII+Berkeley        1 MB memory
			1 MB floppy
			15MB winch

Altos 68000   $12990    8 MHz 68000     24.9  5.5  8.8  10.2  9.8  0.1  24.2
Xenix                   20 MB winch
			512KB floppy
			16 serial ports
			512K memory

Pixel         ~15-18K   8 MHz 68000     35.0  4.9  6.4  21.0 12.6  0.3  24.2
V7+Berkeley             1 wait state
extensions              512K memory
			20 MB winch
			2-1MB floppies
			4 serial ports

Plexus P/25   $16900    Z8000           24.6  3.7  3.4  17.2 17.1  0.1  24.3
SIII+PWB                (speed n/a)
			512KB memory
			32MB winch
			DCD tape

Fortune 32:16 ~$10000   5.5 MHz 68000   19.0  3.2  7.9  15.0 14.0  0.1  25.2
V7 (unbundled)+         5MB winch
extensions              800KB floppy

Onyx C8002A   $15490    6 MHz Z8000     16.8  2.8  5.9  16.9 16.6  0.1  25.4
SIII                    (speed n/a)
			256K memory
			5 serial ports
			20 MB winch
			DCD tape drive

TRS Model 16  ~$10000   68000           24.0  3.2  6.0  17.0 16.1  0.3  25.6
Xenix                   (speed n/a)
			up to 3 users
			12 MB winch
			1 MB floppy
			256K memory

Apple Lisa    ~$10000   5 MHz 68000     60.0  6.1  7.6  16.0 12.1  0.1  25.9
Xenix                   1 MB memory
			5 MB winch
			1 MB floppy

Apple Lisa    ~$10000+  5 MHz 68000     39.0  5.3  7.2  15.0 14.5  0.1  27.1
Unisoft       cost of   1 MB memory
	      second    2-5 MB winch
	      drive     1 MB floppy
	      (n/a)

Dual          $16660    8 MHz 68000     21.0  5.9  6.1  18.0 17.2  0.2  29.4
Unisoft                 (10MHz planned)
(reconfiguration        S100 bus
rights cost $2000)      20 MB winch
			1 MB floppy
			512K memory
			4 users

BBN C/60      $53000    8 users          9.0  1.4  1.9  27.0 26.0  0.1  29.4
BBN Unix                68MB disk
			512KB memory
			The salesman
			said this machine
			was running with
			C/70 microcode.

Momentum Hawk $18000    6 MHz 68000     33.0  4.2 12.6  16.0 12.2  1.6  30.6
model 32/4              (8 MHz planned)
V7+Berkeley             1 MB memory
extensions+f77          20 MB winch
			4 terminals

IBM PC+Sritek ~$8000    8 MHz 68000     38.0  8.9 14.8  11.0 10.6  0.2  34.5
68000 card              1 wait state
Xenix                   Davong 5 MB winch
			2 floppies
			512KB memory

Intel 86/330X $16000    8 MHz 8086      16.0  3.0  4.8  37.0 28.6  0.2  36.6
Xenix                   Multibus
			2 users
			384 KB memory
			35 MB winch

Cosmos Lyra   $16500    68000           34.0  7.9  8.2  23.0 21.7  0.2  38.0
V7 (SIII planned)       (speed n/a)
			20 MB winch
			1 MB floppy
			Multibus
			512K memory
			8 serial ports

IBM PC        ~$6000    5 MHz 8088      26.0 (est.)     26.8 26.0  0.5  39.5
Mark Williams'          512K memory                                     (est.)
Coherent                Corvus 10 MB winch
			IBM monochrome crt

National NS16000 n/a    6 MHz NS16032   37.0  4.4 15.5  21.1 19.6  0.6  40.1
4.1BSD                  1.5MB memory
			20 MB winch
			DCD tape drive
			joystick and
			graphics display

PDP 11/23Plus   n/a     11/23Plus cpu   26.0  3.3  8.3  30.0 29.2  0.2  41.0
DEC V7M-11              2-RL02s
(~$1K per user          size of memory n/a
license fee)

Corvus Unixplex ~$13-14K  8 MHz 68000   40.3  3.5 35.0  14.2 12.5  1.6  52.6
Unisoft         OEMs only 512KB memory
			  2 serial ports
			  10 MB winch


We would appreciate any additional information or corrections to this list.


	Dave Emberson                        Yin Shih
	Megatest Corp.                       Megatest Corp.
	477 Potrero Rd.                      477 Potrero Rd.
	Sunnyvale, CA 94086                  Sunnyvale, CA 94086
	ucbvax!lbl-csam!megatest!dre         ucbvax!lbl-csam!megatest!yin

			  SCO's Case Against IBM

November 12, 2003 - Jed Boal from Eyewitness News KSL 5 TV provides an
overview on SCO's case against IBM. Darl McBride, SCO's president and CEO,
talks about the lawsuit's impact and attacks. Jason Holt, student and 
Linux user, talks about the benefits of code availability and the merits 
of the SCO vs IBM lawsuit. See SCO vs IBM.

Note: The materials and information included in these Web pages are not to
be used for any other purpose other than private study, research, review
or criticism.