Benchmarks
==========
When I saw Logout's post about Windows NT 3.1 [1] and software
non-compatibility then I have tried to find what version on the ANSYS
[2] I saw at the BUT in Brno in late nineties. I never used NT 3.x in
person (my first NT at the BUT was the 4.0) and the firs ANSYS that I
have used was the 5.3.
But I found this nice marketing document [4] in which the ANSYS 5.3 on
the Windows NT 3.51 is cited. Thus I assume that at the BUT it was the
ANSYS 5.x what they were used i that time.
I also studied the tables and graphs inside the [4]. There is a lot of
comparison of PentiumPro-based machines with their UNIX counterparts.
As I have been a SGI used for a long time, I was curious about
PentiumPro-to-MIPS comparisons. And they are non a fair one in this
paper (of course?). The poor low-end SGI Indy with R5000 CPU (that one
which is optimised for single-precision float point computations and
not for double-precision computations) is listed in most of CPU
performance tables. There is almost no comparison with machines
equipped with more powerful R10000 CPUs (or at least with those with
older R4400 ones). It is nice that PentiumPro is 2x faster than R5000
but R10000 should be 2x-4x faster than the R5000. It is the reason why
it is absent here?
First I thought that they wanted to compare their PentiumPro product
with low-end workstations only. But in the graphics performance tables
there are named Indigo2 machines with R4400 CPU and even with R10000!
In some places one even can find the rare POWER Indigo2 with its
multi-core CPU (it does not perform well in the particular test where
it is shown - but it may be fair as its unusual CPU required specially
tuned code to use full power of that CPU).
I remember a case of one of my colleagues at the BUT. He wrote a finite
element method code and ran in on its Intel desktop (I think that it
had the Pentium II, not the older Pro) with the Linux OS. When he ran
the same code on a SGI Origin 2000 computer then the execution time was
seven time shorter. Of course, if was also influenced by the difference
between the GCC compiler on the Linux and the MIPSpro compiler on the
SGI (I assume that at that time the MIPSpro produced better code which
was very well optimized for the R10000 CPU). Anyway, the R10000 ran on
195 MHz and the Pentium was over 200 MHz - so the frequencies were
comparable - and the "slower" CPU was many times faster. But the [4]
tries to convince the reader that it is not the case ;-)
OK, all this makes little sense today. But I have somewhat nostalgic
mood now. I should run the ANSYS tomorrow to compute something...
References:
[1]
http://technomorous.eu/post/165483206330
[2] http:///www.ansys.com
[3]
https://www.fce.vutbr.cz/en_2014/
[4]
http://datasheets.chipdb.org/Intel/x86/Pentium%20Pro/24299903.pdf