[gmx-users] It appears that the Claim that NAMD works with gromacs is false

[Jim Kress]

Thanks David,

> NAMD scaling is great, largely because the single processor code is so
> slow. Try running your system on a single processor with gromacs and on
> four or eight with NAMD and compare the performance.
>
> I can't help you with your question though, sorry about that. I wasn't
> even aware of the possibility.

That may be somewhat the case.  However, I believe the scaling is so good
because NAMD uses a lower level, socket approach to interacting with the TCP
network.  I've seen a similar approach used with other packages (like
GAMESS) which also scale quite well.

The best of both worlds (at least for those of use who can't afford Myrinet)
would be for gromacs single node speed and a sockets TCP networking to get
good scaling.

Jim

PS

Apparently, the NAMD people use the gromacs compatibility as a marketing
point ...