Lies, Damned Lies and...
========================

I have  been extending my  FEA (Finite Element Analysis)  software [1].
One of the  things I have been working  on is a 3D FEA  module). I have
several specific uses for it but it's probably not interesting here.

It is  written in Tcl/Tk  and is compatible  with Tcl/Tk 8.0  to ensure
smooth work on  my SGI machines. There are some  C libraries (mostly to
speedup  computations). Nothing  special. I  confess that  I use  AI to
helping  me with  this (it's  actually useful  as I  was newer  good in
debugging and in searching in libraries documentation).

The  basic package  is a  for 2D  beam-only problems.  Quite sufficient
for  most of  civil  engineering. The  3D package  is  for solution  of
problematic details  and so on.  The 2D thing  works quite good  and in
acceptable speed even  on a 100 MHz  MIPS R4000 CPU (it  is not exactly
fast for tasks like stability and dynamics, of course).

The 3D  part is of  course a tougher nut  for older machines.  It's not
slow but of course modern machines can run the code much faster. I have
been using a C-based  FEA package on the SGI O2 for  two decades so I'm
used to wait. This is even a bit slower as some computations are always
done by the Tcl interpreter. 

What is quite nice  is the fact that I'm now able to  use OpenGL in the
Tcl window so I can have accelerated graphics. My older C code is a bit
heavy and complicated and the  OpenGL performance was always limited by
other parts  of code. Here  things are better  as the data  handling is
easier  and faster  and models  are simpler.  So I  finally can  easily
manipulate models. Even on the 1998' SGI  O2. By the way, I use GLX for
that so  accelerated rendering  works exclusively on  the IRIX  and the
Linux (and  possibly on other systems  which use X11 or  XWayland). The
Windows and MaOS machines use Tcl canvas (which is muuuch slower...).
Now back to the title: I have got an idea that it might be nice to have
a benchmark suite in this 3D FEA.  So I designed some problems to solve
an tried to run them on different machines.

The SGI O2 has score of 60  points, the MNT Pocket Reform can reach 600
(if  fast cores  are  not blocked  by  something else...).However,  the
Lenovo x51s (a 2006 machine, if  I'm not mistaken) reaches 500! (all of
this is for computations as  the overall GUI experience is subjectively
much worse).

My Windows  machine at work reaches  some 1200 points (must  check what
hardware it actually is). Waht surprised  me (because I ignore news) is
that the  MaCbook Neo  (yes, taht  thing with  phone CPU)  reaches 1000
points.

The Pocket  Reform score was  a bit confusing.  It might be  related to
power daemon,  I need to  check. I already  noticed that the  "big" MNT
Reform 2 - with  the same CPU and RAM - can be  2 times faster than the
Pocket. I will try.

But remember,  there are  three types  of lies:  lies, damned  lies and
benchmarks.

P.S. I have  to upload updated version. So don't  play with the current
version now, you will be disappointed.


Written on the SGI O2.


References:

[1] http://jirka.sdf.org