E8D@PSUVM.BITNET (07/11/86)
Here are some benchmarks that I, for one, found illuminating. They
are summarized from The MathWorks Newsletter v.1 #1. who I'm sure won't
mind since they are run on PC-MATLAB, a matrix computation package avilable
from (you guessed it): The MathWorks Inc.
158 Woodland St.
Sherborn, MA 01770 (617)653-1415
The gist of the tests is that this package which performs floating-
point intensive calculations runs at about the same speed on an AT as on
a PC since the 80287 coprocessor on the AT runs at about the same 4.7MHz
as the 8087 on the PC. The ATT 6300 and the Compaq Deskpro, however, are
faster due to the 8086 and 8087 running at 8MHz. "(Note: there may be some
faster AT clones.)"
The test results as ratios to the PC times are:
| PC (sec)| AT:PC | ATT 6300:PC | SUN 3/150:PC
| | $4500 | $2000 | $10,000(12.5MHz)
------------------------------------------------------------------------
1)Multiplication of | 10.0 | 0.87 | 1.8 | 2.4
two 50x50 matrices | | | |
2)Inversion of a | 23.5 | 1.0 | 1.9 | 5.2
50x50 matrix | | | |
3)Eigenvalues of a | 24.4 | 1.1 | 1.7 | 6.0
25x25 matrix | | | |
4)FFT of a 4096 point | 16.0 | 1.0 | 1.8 | 3.9
vector | | | |
5)Linpack benchmark | 47.1 | 0.9 | 1.8 | 4.6
6)1000 iteration | 29.4 | 3.1 | 2.1 | 13.8
for loop | | | |
7)25x25 element 3-d | 16.8 | 1.7 | 2.1 | n.a.
mesh plot | | | |
--------------------------------------------------------------------------
The article notes that the AT is 2-3 times faster than the PC for
integer and pointer arithmetic in for loops, the parser, and graphics.
The ATT is faster mainly because of the clock speed, but also for integer
arithmetic (benchmarks 6 and 7) because of the 16 bit data/address path of
the 8086 instead of the 8 bits for the PC's 8088.
Any opinions on this as an assessment of machines for someone who
wants to do a lot of number-crunching?
Note: I don't have anything to do with MathWorks and haven't even
compared their COPY PROTECTED package on my 6300 to my advisor's AT. As
a matter of fact I probably won't bother running down the hall enough
to grow attached to anything that I can't make real backup
copies of, so I doubt I'll recomend it's purchase to anyone now or
to any future employers.
-------
Disclaimer: Sorry about the spelling.
Evan Dresel
Dept. of Geochemistry E8D @ PSUVM (bitnet)
228 Deike Bldg. ...!psuvax1!psuvm.bitnet!e8d (uucp <-->
Penn State University bitnet gateway)
University Park, PA 16802 e8d%psuvm.bitnet@wiscvm.arpa (arpa)
(814) 863-0672