[Help-glpk] Hardware recommendations for GLPK

[cgi2]

We're in the process of selecting hardware to run GLPK on. We'll be solving LPs 
with hundreds of thousands of
variables. Does anyone have recommendations as to what hardware would be best 
suited to it. We're mostly considering
x86 hardware but may go for Sun or IBM hardware if they're significantly faster.

Last year I'd compared binaries compiled with 1) Microsoft's VC6 compiler and 
2) Intel's v7 compiler on a typical
problem and found the latter to be 2.6 times as fast. This was on a 2.4GHz P4 
without doing any processor specific
optimizations. Doing processor specific optimizations (SSE2) made no difference 
in performance. I also tried the Intel
compiler directive to try and automatically parallelize it. However the 
compiler failed to find any place where
parallelization could help.  So it looks like multi cpu machines wouldn't offer 
any significant advantage.

Since Sun and IBM compilers are unknown quantities to me, I'm inclined to 
choose an x86 machine so I may continue to
use the Intel compiler.

Is there any advantage to going to a larger cache machine like Xeon or Opteron? 
 Which would be faster: the Xeon or
Opteron?

How are Itaniums with GLPK?

Which solution would you recommend if money were no object (within reason)?

Are there any benchmark comparisons between GLPK and leading commercial solvers 
like CPLEX on large LPs?

Thanks
CG
________________________________________________
Don't E-Mail, ZipMail! http://www.zipmail.com/