From: Mert Gür (gurmert_at_gmail.com)
Date: Fri Oct 19 2012 - 08:49:43 CDT
I think during the NAMD workshop in 2010 this issue was discussed.
However, I do not remember the details.
My understanding was that NAMD scales well up to 1000 atoms per CPU. After
that (like 500 atoms per CPU) the scaling does not go linearly.
Hope it was of some help.
On Fri, Oct 19, 2012 at 9:32 AM, Isaure Chauvot de Beauchêne <
> I plane to run a parallel MD simulation of a 6500 aas protein in gb
> implicite solvant on the cluster describe below**. I need to compute an
> estimation of the CPU time necessary before I get access to the cluster
> where it will run.
> I did a benchmark on 8 CPU, but I couln't find anywhere benchmarks about
> the paralleling efficiency of NAMD for such a small system.
> How could I get an estimation of the optimal number of 8-CPU nodes to use
> (other than tests, since I don't have access to any cluster now)? I only
> need a rough approximation.
> **The Cluster:
> 1536 compute nodes i.e. 12288 cores of Harpertown processor and 1344
> compute nodes of nehalem processor : 10 752 cores. The network is an
> Infiniband (IB 4x DDR) double plane network for the Harpertown section of
> the machine. On the Nehalem section of the machine, 4 InfiniBand 4X QDR
> drivers provide 72 IB 4X QDR ports on output of each IRU (576 Go/s).
> Many thanks in advance,
> Dr Isaure Chauvot de Beauchêne
> Ph.D student
> Team: "Bioinformatics, Modelisation and Molecular Dynamics" (BiMoDyM), Dir
> L. Tchertanov.
> Laboratory of Biology and Applied Pharmacology (LBPA)
> CNRS UMR 8113 - ENS Cachan
> Bâtiment d'Alembert, 3e étage
> 61, avenue du Président Wilson
> 94235 Cachan cedex
> Phone: +33 1 47 40 68 76
> Fax: +33 1 47 40 76 71
This archive was generated by hypermail 2.1.6 : Tue Dec 31 2013 - 23:22:40 CST