|Clock freq. (MHz)
|Gemini quoted value ‘about 6MHz’.
|Results obtained in earlier
(see Issue 1)
The results are of considerable interest since they show that the quoted speed
of the Gemini 8 bit CPU and 8/16 bit co-processor combination is much greater
than the actual value [Ed. – NO! See my comments below.] The Amstrad, working
at 8 MHz is actually 1.8 times faster than the Gemini combination which appears
to have a true overall speed of 4.44 MHz, marginally slower than the 4.77 MHz of
the IBM PC. I was a little concerned about this so tried the program on a
Samurai 16 bit machine (which runs MSDOS and CP/M-86) using the MSDOS and CP/M 86
versions of BASIC. The system clock is 4.608 MHz and the CPU is an 8086.
The numerical values obtained were the same as before but timings were 115 secs
for the MSDOS version and 122 secs for the CP/M-86 version, thus confirming the
slowness of the Gemini combination.
[Ed. – sorry, but there is one factor here that you appear to be omitting to
take into account, although you did mention it earlier. The GM888 and IBM-PC
both use the 8088 processor which only has an 8-bit external data bus, as
opposed to your Amstrad which has an 8086 with 16-bit data bus. It is therefore
not correct to assume that the IBM PC speed would be 4.77/8 times that of the
Amstrad, it is considerably slower than this. Nor is it correct to calculate
the effective Gemini clock speed by comparing its benchmark result to that of
the Amstrad. Neither of these calculations are comparing like with like.
When the GM888 board was initially introduced Gemini ran benchmarks on it and on
the 4.77 MHz IBM-PC. The GM888/MultiBoard combination ran the benchmarks
approximately 25% faster than the IBM – hence the “effective” speed of the
Gemini is about 6MHz, as quoted.
The areas where your figures are obviously interesting are (a) in being able to
calculate the performance improvement to be had from using the 16-bit data bus
versions of the Intel family as opposed to the 8 bit ones, and (b) in seeing the
ACTUAL performance differences of the Amstrad vs. the Gemini.]
Microsoft appear to have improved the accuracy of their BASIC in it’s MSDOS
version and out of interest, I looked at the performance of their 16 bit FORTRAN
and PASCAL which run under MSDOS. Using appropriate versions of the benchmark
program, Fortran-77 ran in 162 secs. on the Amstrad and produced an acceptable
value of 2477.24 while the Pascal completed in 130 seconds and produced an
appalling 3904.77. I’m glad I don’t have to use it! I had hoped to try out
the MSDOS version of ProFortran and the CP/M-86 version of ProPascal but I
didn’t have access to copies. I suspect that they would run fast and produce
reasonably accurate results.
I included both versions of Ellis Computing’s FORTRAN because they are cheap and
quite accurate. Their execution speed is a little slow but for many
applications, this is not a great disadvantage.