mjg@ecsvax.UUCP (Michael Gingell) (03/06/86)
In an effort to speed up my PC I have been looking at 80286
coprocessor boards and the use of the 8087 or 80287 math coprocessor.
Last week I posted a request for information on the performance of
80286 boards, so far I have had 2 replies. When I get enough to
make sense I will post a summary.
In the mean time I have come up with some interesting results re
the use of 8087 chips which many people may not be aware of. I used
the Dhrystone and Whetstone Benchmarks compiled with Microsoft C V3.0
on a regular PC, an XT with an 8087, and an ITT Xtra XP with 80286/7.
The Xtra is claimed to be XT compatible and run 30% faster than an AT.
Results were as follows :
Dhrystones/sec:
Compiler model : small medium large small 80286
PC, no 8087 335 297 265 -
XT, with 8087 333 296 264 -
Xtra with 80287 1250 1111 1063 1351
so far the Xtra with 80286 lives up to its claims and the 8087/80287
makes no difference because the Dhrystone test has no floating point.
Whetstone test:
Note: The MS compiler has 3 options
a) Compile with conditional code which executes on the 8087 if present
or an 8087 software emulator if not.
b) Compile with in line 8087 code. This is fastest.
c) Compile with alternate math library. This is fastest if you don't
have a math co-processor.
Any of the above can be selected while compiling for 8086, 80186 or 286
also you can choose small, medium or large models. I used small.
Whetstones/Second:
Option PC no 8087 XT with 8087 Xtra with 80287
8086 a) 0.0511 1.3333 1.4493
8086 b) - 1.4706 1.5152
8086 c) 0.0929 0.0929 0.3937
80286 a) - - 1.4925
80286 b) - - 1.4286
80286 c) - - 0.3937
The results show that when the 8087 is not used the Xtra is about
4 times faster than the XT or PC. However the 8087 speeds up the
XT by about 16 times. The most significant result is that the 80287
on the XP gives no significant advantage over the 8087 in the XT.
Now the question is are any of the co-processor boards any better.
At least some can use an 8 MHz 80287 instead of a 5 MHz one which
would givea 60% speedup.
Immediately following this posting is one with the uuencoded benchmark
files for wet.exe, the 8086 option a) benchmark and dry.exe, the small
model 8086 dhrystone benchmark.
If anyone out there with a coproccessor board will run these and
pass me back the results I will summarize for the net. If you want
the source I can email it to you but I thought it would be better
to use the identical executable code to eliminate possible errors
due to selection of the many different compile options and even
worse, different compilers.
Please email your replies to me, Mike Gingell
.....decvax!mcnc!ecsvax!mjg
P.S.
Thanks in advance to anyone who can augment the sum of knowledge on
this subject.