RE: [gmx-users] benchmark: buying a new cluster

May 2008 10:58:13 +0200 > From: [EMAIL PROTECTED] > To: gmx-users@gromacs.org > Subject: [gmx-users] benchmark: buying a new cluster > > Hi all, > we are going to buy a new cluster and we are in doubt about the frequency to > consider. > Basically, we have the possibility to bu

[gmx-users] benchmark: buying a new cluster

To some extent, it depends on your interconnect and the number of atoms in the simulations that you will be doing. If you are using gigabit ethernet, then you may not scale very well beyond 8 cores on these dual node quad cores and so there is an argument to be made for 10% faster. However, if you

Re: [gmx-users] benchmark: buying a new cluster

There two CPUs differing not only in frequency but also in FSB speed. But the differences in FSB shall be ignorable for Gromacs. You have two dimensions to think. 1. power of all nodes = number of cores * frequency 2. power of a single node. If you intend to have one simulation running just on

[gmx-users] benchmark: buying a new cluster

Hi all, we are going to buy a new cluster and we are in doubt about the frequency to consider. Basically, we have the possibility to buy either 18 nodes dual quad-core (18x8 cores) bearing Intel-Xeon E5430 2.66 MHz and FSB 1333 MHz or 15 nodes dual quad-core (15x8 cores) bearing Intel-Xeon E5462