Re: About charm++/namd on AMD64 cluster

From: Gengbin Zheng (
Date: Mon Jan 03 2005 - 14:46:58 CST


Adding CPUs sure can slow down program if
1. your atom system is too small, and then imaging more people do one
trivial task can be slower than just one man do the same task alone due
to the synchronization overhead.
2. The communication hardware is slow, for example 100Mbps ethernet is
inefficient especially for small job. In which case, there is not enough
computation work to overlap the communication latency.

so try some larger system for example apoa1 benchmark (you can download
it from NAMD website) and see how it scales.


王小清 wrote:

> I downloaded the latest Charm++ and compiled the mpi
>version of Charm++/NAMD on amd64 cluster.They can work
> I made a simple 100 steps minimization.But why the
>CPUTime is become longer with increasing the number of
>cpus?What factors affect the CPUTime?
>Thanks in advance!
> --- Gengbin Zheng <> 的正文:
>>This errors should be fine as they are fortran
>>related and NAMD does not
>>written in fortran.
>>Have you tried test program in
>>charm/pgms/charm++/megatest? Whatis the
>>If it fails, Maybe you should download the latest
>>source from Charm++
>Do You Yahoo!?

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:40:27 CST