Re: Workstation dedicated for FLUKA

From: Bertrand H. Biritz <beb2015_at_med.cornell.edu>
Date: Wed, 03 Aug 2011 16:35:53 -0400

Hi Tomasz,

The only thing I would add is to understand how to use the CPU most efficiently.

For our simulation cluster (which is all mine at the moment :) we got two
Dell PowerEdge R210 and two R210 II servers. They have an Intel Xeon chip
with TurboBoost technology. The one thing I did not realize at the time
was the Turbo Boost is dependent upon how loaded the system is:

http://en.wikipedia.org/wiki/Intel_Turbo_Boost

If I run with all cores loaded then (at least for the chips I have) there
is no Turbo Boost. If I just have one core loaded then I get the maximum
Turbo Boost.

In the end I found the best balance between CPU time/primary and
simulation completion time being to have half of the cores loaded. This
allows me to complete a simulation of:

66 region geometry
200M primary events
5 cycles with 12.5M per cycle
8 active cores

with a CPU time/primary of 4.7 ms/4.2ms (R210 / R210 II) in roughly 17 to
15 hours (respectively).

Perhaps this was already common knowledge, but just in case it wasn't I
wanted to add that in.

Bertrand

On Aug 3, 2011, at 12:01 PM, Vasilis Vlachoudis wrote:

> Hi Tomasz,
>
> "optimum computer" is a hard to define. Our approach is typically the
> fastest CPU with lots of cache, minimum 1Gb of RAM (a fast one) per
> CPU-core and lots of disk space.
>
> Cheers=0A=
> Vasilis Vlachoudis
>
> Dep EN, CERN
> CH-1211 GENEVA 23
> SWITZERLAND
>
> Phone: +41-22 767 9851
> GSM: +41-76 487 4378
> Fax: +41-22 766 9644
>
Received on Thu Aug 04 2011 - 14:16:47 CEST

This archive was generated by hypermail 2.2.0 : Thu Aug 04 2011 - 14:16:47 CEST