A few benchmarks for the Nanonote

Delbert Franz ddf at sonic.net
Tue May 4 17:07:01 EDT 2010


Here are a few benchmarks of the Nanonote comparing it to:

1. Acer Aspire One- Intel Atom at 1.6 GHz and 1 GB RAM. 

2. Gecko Edubook from:

 http://www.norhtec.com/products/gecko/index.html

using an Xcore86 at 1 GHz, a rebrand of the Vortex86MX:

http://en.wikipedia.org/wiki/Vortex86.

It is in a class similar to a Pentium 200. 

The following are all written in Lua and with Lua compiled to 
use double-precision floating point.

A. Simple prime number sieve computing the number of primes 
<= 1,000,000.  Times in seconds.

 Ben          Gecko       Acer 
-----         -----      -----
 11.2         7.4         1.8


B.  Rough translation of N. Tajima's INTPI3.F:

http://serv.apphy.fukui-u.ac.jp/~tajima/bench/intpi3.f

Lua does not use integers but I used the math.floor
function to force divides to produce an "integer".
The output of the Fortran program and  the Lua script agreed.

This was run with n=250 so that a single-precision Lua
would give the correct result (single-precision outcomes
appear below).  Times in seconds.

 Ben          Gecko       Acer 
-----         -----      -----
 25.5          7.5        1.9


C. A quick and dirty recursive adaptive routine loosely 
based on an F program:

http://www.fortran.com/F/ex_integrate.html

It took quite some doing to find a function to integrate
that would give a large enough time on the Acer to be
reliable and yet not result in stack overflow on the 
Ben:)  Times in seconds.  Convergence tolerance 
set at 1 x 10^(-6), again, so that single-precision
would work:)

 Ben          Gecko       Acer 
-----         -----      -----
 22.1          1.3        0.38


D. Single-precision results for the Ben for each test:
(Double-precision results shown again for quick comparison.)

Precision   Test A   Test B   Test C
--------    -----    -----    -----
Single       9,9      17.2     20.9
Double      11.2      25.5     22.1 

Comments on number crunching:

1. Given the power used by the Ben it does quite well.  It is 
quite a bit slower than the others but then its size and 
power requirements are also much smaller.  The Gecko uses
about 5 watts at full load, the Acer probably several 
times that, and I have no measurements for the Ben but
given the battery life I have found, it has to be 
much less than 1 watt, especially when the screen is blank.
The simple-minded, low-order, adaptive integration routine
on the Ben would often finish in 1 second or less for 
a wide range of test integrals.  The one I used for the 
benchmark was forced to be extreme in order to get
a reliable time on the Acer One. 

2. Using double precision for Lua only increases the 
compute times by a modest amount.  The storage space
available is reduced to one-half the single-precision
value. For example, I could run the single precision
version of the sieve to 2,000,000 before I got 
an out of memory failure. 



E. I/O speed via network to and from the Ben.  Connected
using a Debian variant of process taken from here:

http://en.qi-hardware.com/pipermail/discussion/2010-March/000127.html

I used scp for the transfer and each one involved binary data
with multiple large files.  These were *.ogg files I have 
on the Ben.  I am booting from the a microSD card.  It is 
a Transcend 8 GB Class 6 card that retails for about $20 US.
The data is stored on one of the partitions on that card.

  Write speed:  450 KB/sec
  Read speed:   725 KB/sec

The Ben was hosted on the Acer One netbook and the transfer was 
to or from an Intel i7. 

All in all I find the results quite pleasing:)

                   Delbert





 










More information about the discussion mailing list


interactive