Comment: Re:Why? (Score 1) 108
Found the system specs for the test PCs, I wouldn't call them High End PCs except in the sense that the price was high end. The Xeon X5550 system appears to even be underclocked 1GHz below it's normal speed of 2.66/3.06GHz.
Still, it'd take a big cluster of them to equal the performance of the V6. That's worth it right there.
http://www.cs.amherst.edu/ccm/cf14-mcgeoch.pdf
"All software solver tests were carried out on a suite of
seven Lenovo ThinkStation S30 0568 workstations, each containing
one Intel Xeon E5-2609/2.4GHz Quad-Core processor
with 16GB RAM. The operating system was Ubuntu
64-bit 12.04 LTS.
Blackbox runs on a Lenovo d20 workstation containing
two Intel Xeon X5550@1.6GHz Quad-Core processors with
16GB RAM. The operating system is Fedora 15. The number
of hardware samples per main loop iteration was set
to k = 1000 and the stopping rule was set to 107 function
evaluations.
The QA algorithm was run on a hardware chip named
Vesuvius 5 (V5) that contains 439 working qubits.
It is a challenging problem to nd precise, accurate, and
commensurable runtime measurements for these diverse solution
strategies. We adopted the following conventions.
All software runtimes are Unix CPU times in units of seconds.
The Matlab front end started timing immediately
before solver invocation and stopped immediately upon return:
thus the tasks carried out by the front end (including
all I/O) are not included in our time measurements. All software
tests were run on empty systems (with no competing
user processes), measuring one solver on one instance, running
on one core at a time. The Intel hyperthreading option
(which is known to produce timing anomalies) was turned
o. In addition to total CPU times, most tests produced
\history" trace data, by which each solver recorded time and
solution cost whenever a better solution was found."