[Wien] WIEN2k 09 poor benchmarks

[Scott Beardsley]

Peter Blaha wrote:
> WIEN2k09 must not be slower than previous versions.

Hmmm, my test results[1] don't coincide with that statement. Is the 
mpi-benchmarks package[2] not a recommended mpi benchmark any more? Can 
you send your benchmarks of 08 vs 09? I don't see any version numbers on 
this benchmarks page[3].

> In fact there is a small improvement in the parallelization of the LOs 
> in hamilt.F (probably only noticeable with very many nodes, since this 
> small part (1-2 % of hamilt) was not parallelized at all, but becomes 
> important only for more than 64 nodes).

So far, up to 32 cpus (four 8-cpu nodes) I'm not seeing this 
improvement, just the opposite.

> fftw is used ONLY in lapw0. While the speedup is modest, memory 
> requirement for large cases is drastically reduced.

Ya I noticed that. Do you recommend double precision or single precision 
FFTW?

Scott
----------
[1] http://wiki.cse.ucdavis.edu/support:hpc:software:wien2k#results
[2] http://www.wien2k.at/reg_user/benchmark/mpi-benchmark.tar.gz
[3] http://www.wien2k.at/reg_user/benchmark/