Ahem,
Maybe I missed some earlier stuff here. The pmemd.amba tree, while REALLY
mostly a snapshot of pmemd 9 with amoeba support and a few minor fixes, is
intended to be different than pmemd 10. I deemed supporting amoeba within
the changing pmemd context to be too complicated to be justified; amoeba is
basically hopeless from a performance perspective for about the next 100
years, pending some significant hardware breakthroughs. So pmemd.amba is
supposed to be a snapshot to support amoeba for the brave few willing to pay
the price. The pmemd 10 drop is "in progress", but not in the tree yet, and
it probably will not go out with the main release unless the main release
slips a bunch. Basically, I set some ambitious performance goals for myself
over the last six months, and beat my head against a brick wall continuously
trying to meet them before finally having to concede that we are close to as
far as we can go with the combination of fortran 90 + mpi + pme + poor
hardware. NAMD gets about 23% higher peak performance than we do now; they
do it by really fine-grained workload definition and their charm++ tasking
engine which basically allows them to completely overlap computation and
communications throughout the step cycle (as compared to us, where we
exchange coordinates, forces, potential energies and virials, and kinetic
energy and max distance traveled info (for pairlist update decisions) at
four distinct points - so we really end up effectively using about 50-60% of
communications bandwidth at very high scaling). So anyway, the pmemd 10
drop will come into the tree late, and will probably be up to 50% faster
than pmemd 9 for select cases (ntp won't be much faster at all). So I
wouldn't go renaming pmemd.amba to pmemd is I guess my point... The real
pmemd 10 is still coming.
Best Regards - Bob
----- Original Message -----
From: "Andreas Svrcek-Seiler" <svrci.tbi.univie.ac.at>
To: <amber-developers.scripps.edu>
Sent: Tuesday, March 25, 2008 10:47 AM
Subject: amber-developers: pmemd is not built out of the box
> Hi all,
> pmemd(.amba) is not built from the current (Mar. 25) cvs tarball.
> This comes from mixing the names pmemd and pmemd.amba.
>
> *One* solution:
> cd $AMBERHOME/src/pmemd.amba/src
> cp pmemd.fpp pmemd.amba.fpp
>
> The resulting pmemd.amba should then be renamed to pmemd,
> since alle the scripts in the $AMBERHOME/test expext
> 'pmemd' rather than 'pmemd.amba'.
>
> I hope this is not a duplicate :-)
>
> best
> Andreas
>
>
>
> )))))
> (((((
> ( O O )
> -------oOOO--(_)--OOOo-----------------------------------------------------
> o Wolfgang Andreas Svrcek-Seiler
> o (godzilla)
> svrci.tbi.univie.ac.at
> .oooO Tel.:01-4277-52733
> ( ) Oooo.
> -------\
> ---( )--------------------------------------------------------
> \_) ) /
> (_/
>
>
>
Received on Fri Apr 18 2008 - 21:13:56 PDT
