[Beowulf] 64-bit AMD Opteron with 32-bit Intel Xeon possible?

Mark Hahn hahn at mcmaster.ca
Fri Aug 31 12:46:27 PDT 2007


> For now I have only 2 dual-processor (smp) machines from ASL that I am trying 
> to get working together with a fortran 90 code that includes the LAPACK 
> solver ZGESVX.  One machine has 32-bit Xeons, the other has 64-bit Opterons.

I would run a 32b environment on both and ignore the nice features
(registers) that come along with the opterons.

> 1.)  Is it recommended to compile a single executable that will run on both 
> machines as I have done, or is it easy to allow different executables (i.e. 
> make a 64bit executable for the Opteron machine)?

you should evaluate the possible gain from customized systems/binaries.
first, the premise is that 64b binaries will run more efficiently, 
presumably due to being able to address more memory, or because they 
have more/wider registers.  for your code, neither might be true.
further, you need to evaluate whether having a subset of nodes run faster
will benefit the overall performance, since to do so would require 
some sort of work adaptation (finer-grained work queue, adaptive 
division of work, etc.)

I find that most people, once they choose to invest in MPI, tend to just 
toss another handful of cpus on the fire, rather than struggling to tune
that extra 5% by sweating the 64b-ness, etc.

> 2.)  When installing MPICH1, should I configure/make/make install separately 
> on both machines, or will I end up with library incompatibilities because of 
> the 32 vs. 64bit architectures?  Should I just install MPICH1 on one machine 
> and copy over the resulting directory tree /usr/local/mpich?

I think you should run both as xeons and simply nfs share such packages.



More information about the Beowulf mailing list