I briefly tried to compare computational speed of my new Gemini PDA
with my older GPD Pocket. I used by own uFEM finite element solver [1]
for that.
The GPD Pocket runs Ubuntu 16.04 on an quad-core Atom CPU. The Gemini
PDA runs Android (7.1, I think) on an octa-core ARM64 processor. I used
the default GCC on the Ubuntu and the default GCC which was a part of
the Termux program for Android.
Of course, I am aware about the fact that the Atom has four identical
cores but the ARM includes several types of cores with several speeds.
The uFEM is not the best example of use of symmetric multithreading, of
course So I was curious about results.
The results were somewhat interesting for me:
If one core is used then the speed is comparable: the GPD/Intel needs
74 seconds and the Gemini/ARM needs 72 seconds.
The best time is 47 seconds for the GPD and 46 seconds for the Gemini.
It was obtained with 4 cores on the GPD but with only 3 cores on the
Gemini. It is interesting that use of 4 cores on the Gemini gives worse
time (54 seconds).
This is interesting for me because it is said that the GPD has 1.6GHz
CPU (2.67 GHz in full speed) with four identical cores and the Gemini
has 2x2.6GHz cores and 2x2.0GHz cores (and a handful of slower ones).
I have repeated program runs several times, of course.
By the way, my old (no longer working) Opteron-based system (the Sun
Ultra 20 with 2.2GHz amd64 CPU) needed 173 second and my current main
desktop workstation (the SGI O2 with 250 MHz CPU) needs 2750 seconds to
solve the same problem. Of course, this O2 is 20 years old so there
should be some speed difference. ;-)
