zemon@felix.UUCP (Art Zemon) (04/14/87)
Earlier I posted a couple of Dhrystone benchmark results
and failed to compare them to anything. My tone in that
article came across as a bit too negative so please let me
try again.
I ran the Dhrystone benchmark on a couple of Minix machines
with the following results. When you read the table, keep
these points in mind:
1) This benchmark tests the CPU/compiler combination
for a "average" mix of instructions. This mix
includes integer arithmetic, branches, etc.
2) There are no floating point operations in the
Dhrystone program. (A good thing, since floating
point is not implemented in the Minix C compiler.)
3) Like most benchmarks, the numbers which come out of
this one are not gospel. You can compare
Dhrystones between two different machine/compiler
combinations and make some sort of judgement about
relative speed, but they do NOT tell the whole
story. Read point 4.
4) The Dhrystone does not test any operating system or
hardware functions. It is completely CPU bound in user
mode -- no system calls, no I/O.
So with that said, here are the results I came up with. My
Zenith is an XT clone with a V20 chip running at the
standard XT clock rate. The V20 seems to run programs
about 5% - 10% faster than an 8088 in MY experience. The
AT is a "real" AT with no mods at all.
DHRYSTONE 1.1 BENCHMARK SUMMARY
MANUF MODEL PROC CLOCK NOREG REG OS,COMPILER,NOTES
----- ----- ---- ----- ----- --- -----------------
Zenith 151 NEC V20 4.77 162 Minix, cc
IBM PC/AT 80286 8.00 767 Minix, cc
DEC VAX-11/750 w/FPA 0.00 831 852 UNIX 4.2BSD, cc
IBM PC/AT 80286 9.05 1464 1484 XENIX SCO SVR2.1,cc small,
IBM PC/AT 80286 8.00 1729 1796 PC-DOS 3.20,Microsoft 4.0 ,
IBM PC/XT 8088 4.77 326 347 MS-DOS 2.0,Microsoft 3.01 ,
If you have read this far, you might as well wade through
my personal observations on all these numbers. Thanks to
Andy T. for pointing out that the AT compares very
favorably to a VAX 750 and starting this monolog off.
The XT is pretty darned slow. The Dhrystone runs about
half as fast as the old Microsoft C compiler on the same
CPU and that was not a very good compiler. On the other
hand, I can sit at my Zenith and type "ls" and various
other commands and the response feels about the same as the
VAX 785 I share with a bunch of other software developers.
So although the V20/Minix compiler combination is slow,
there is a lot to be said for having the CPU available
RIGHT NOW, when I need it, and ALL to my self. But if I
had serious number crunching to do, this would not be my
system of choice. One option is to compile such a program
with a "good" compiler under MS-DOS and then convert the
.EXE file to a.out format. Another is simply to do my
number crunching under MS-DOS. Or maybe somebody with the
compiler source will clean up the emited code....
The AT, on the other hand, is very fast for a single user
machine. It runs this benchmark at about 90% the speed of
a VAX 750. Definitely a nice machine for a single user.
Using it "feels" like you have a 750 to yourself -- until
you try to do disk I/O. Here you run into one of the
things which the Dhrystone does not test. The 750, while
not much faster than the AT for Dhrystone cycles, has a lot
of I/O hardware. I suspect that you will get two
simultaneous comilations done much more quickly on the 750
than on the AT. But then again, the AT is your personal
single user machine.
All in all, I'm pretty happy with the performance on my
Zenith. It isn't very fast but all I'm doing right now is
hacking and I'm not in any hurry to "get it done." (I
reserve that attitude for the workday!) If I had an AT I
could easily use that for my main computer.
--
-- Art Zemon
FileNet Corporation
Costa Mesa, California
...!hplabs!felix!zemon