info-mac@utcsrgv.UUCP (info-mac) (06/12/84)
Date: Sun, 10 Jun 84 17:54:20 CDT
From: Don Johnson <uw-beaver!dhj@rice.ARPA>
To: info-mac@sumex-aim.ARPA
Subject: Benchmarks for Macintosh and Lisa
Cc: dhj@rice.ARPA
As part of an extensive set of benchmarks, the Macintosh and Lisa
machines were subjected to study. These benchmarks differ significantly
from the others in several important respects. The fundamental
difference in this benchmark procedure was that the language used was
PASCAL while the others were written in C. Furthermore, a comparison
the floating point results is difficult. The Macintosh and Lisa are
based on the Motorola 68000 and have no hardware support for
floating-point computations. The floating-point software package
provided by Apple Computer, SANE (Standard Apple Numerical
Environment), does NOT support computations with the standard formats
of single and double precision. Rather, an extended format allocating
80 bits per floating point word is used. Thus, more precision is
obtained than in all other benchmarks. The results are summarized
below; the numbers indicate the time taken for each high-level language
instruction. The numbers in parentheses are the performance ratios: the
ratio of the time taken for a standard computer to that taken by the
machine of interest. Thus, a computer taking half the time of the
standard to execute a given expression would have a performance ratio
of two for that operation. The standard used here was a Texas
Instruments Professional Computer without 8087 support. Note that the
TI-PC is essentially the same as an IBM PC but with a slightly faster
clock (5 MHz versus 4.77).
Personal Computer Benchmarks
TEST TI-PC TI-PC Macintosh Lisa
w/8087 MacWorks
16-bit Integer Operations (USEC)
J=I+K 13.5 5.6 (2.41) 7.2 (1.88)
J=I*K 49.0 11.4 (4.30) 16.0 (3.06)
J=I/K 57.0 26.7 (2.13) 37.7 (1.51)
32-bit Integer Operations (USEC)
J=I+K 50.0 7.6 (6.62) 8.1 (6.17)
J=I*K 184.0 75.5 (2.44) 86.9 (2.12)
J=I/K 1282.0 438.6 (2.92) 526.7 (2.43)
Floating-point Operations (USEC)
X=Y+Z 859.5 282.5 (3.04) 536.1 (1.67) 601.2 (1.43)
X=Y*Z 1469.0 286.0 (5.14) 527.7 (2.78) 606.2 (2.42)
X=Y/Z 9179.0 305.5 (30.05) 1374.4 (6.68) 1696.7 (5.41)
Computational Benchmarks (MSEC)
SQRT 50.0 4.6 (10.87) 2.1 (23.4) 2.8 (18.0)
SIN 31.8 8.8 (3.61) 15.6 (2.04) 18.3 (1.73)
EXP 36.2 6.8 (5.32) 20.5 (1.77) 24.2 (1.50)
LOG 60.5 6.7 (9.03) 22.7 (2.67) 26.9 (2.25)
On the average, the Macintosh was 1.21 times faster than the Lisa
(implying a virtual clock speed of 6.07 MHz). Also note that the SANE
software has a VERY fast square root routine according to these
measurements (only 0.15 that of a VAX 750 with a floating point
accelerator!!).
If you are interested in the full report on my benchmark study,
just ask. If the demand is small, it shouldn't take long for you to
receive your copy. However, note that it focusses on minicomputers
more than on micros.
Don Johnson
EE Department
Rice University
dhj@rice