Clovertown HPL scores

[Kilian CAVALOTTI]

On Tuesday 27 March 2007 05:00:32 Peter Kjellstrom wrote:
> huh? mvapich is an mpi-implementation for IB. The OP did not mention
> using infiniband. Also for Linpack I know Sandia used OpenMPI on
> Thunderbird to improve on their first figures achieved with mvapich.

Actually, our cluster uses an Infiniband interconnect,  but it's not 
relevant yet, since I just want to run tests locally, get the most 
efficiency out of one node. I guess that the efficiency will decrease 
again when using inter-nodes communication, so I'd prefer have the best 
starting point before scaling. I may be wrong, but I guess for local jobs, 
the MPI implementation won't be that significant.

I was rather looking for a way to determine HPL.dat parameters in a 
rational way. Playing around with the settings is not really easy when you 
don't know if the parameters are correlated, and if what direction you 
should go. For instance, are N and NB related in a way, or are they 
totally independent?

Thanks,
-- 
Kilian