From: Bill Knebel billk@metrumrg.com
Subject: [NMusers] processor choice
Date: Mon, 21 Nov 2005 16:18:41 -0500

A few years ago there was a thread about benchmarking results for NONMEM across a number
of different processors and systems (http://cognigencorp.com/nonmem/nm/98apr292002.html). 
Given the relatively recent release of dual core processors I wanted to know if anybody
has compared NONMEM runs on a dual processor machine to NONMEM runs on a dual core CPU. 
Assumming you were running two jobs simultaneously, I would be interested if anybody has
seen any speed (runtime) increase or decrease with either setup.
Bill

-- 
Bill Knebel, PharmD, Ph.D.
Principal Scientist
Metrum Research Group
2 Tunxis Road
Suite 112
Tariffville, CT 06081
email: billk@metrumrg.com
tel: (860) 930-1370 

_______________________________________________________

From: Darin Perusich Darin.Perusich@cognigencorp.com
Subject: Re: [NMusers] processor choice
Date: Tue, 22 Nov 2005 08:41:48 -0500

for an application to take advangate of a dual core cpu, much like pentium4's
hyper-threading it needs to be a multithreaded and nonmem is about the furtherest
thing from multithreaded.

i did some runtime comparisons on single cpu P4 systems with HT (hyper-threading) enabled
and disable, and dual cpu P4 systems with HT disabled. HT'ing doesn't by you anything when
running nonmem other then a longer runtimes when you try to run a job/cpu. take a job that
runs in 40 minutes single cpu with HT disable. when HT is enable and 2 identical jobs are
run, one per virtual cpu the jobs take nearly twice as long. though runtime is marginally
better with HT then when running 2 jobs when HT is disabled. when run on a dual-cpu system
with HT disabled of the same speed, the 2 jobs run in the same time as a single job when HT is disabled.

i can purchase a machine with dual 3.0GHz Intel Xeon processors for $1779.00, a dual-core
3.0GHz intel pentium D runs about $1707.00, or a singe 3.0GHz P4 for around $1550.00. if
your goal is the maximize the number of nonmem jobs you can run in the shortest amount of
time your better for going with dual processor machines any day.

my 2 cents.

-- 
Darin Perusich
Unix Systems Administrator
Cognigen Corp.
darinper@cognigencorp.com 

_______________________________________________________

From: mark.e.sale@gsk.com
Subject: Re: [NMusers] processor choice
Date: Tue, 22 Nov 2005 09:14:45 -0500

Darin is correct, NONMEM is single threaded.  Even if it were multi 
threaded, it probably couldn't take advantage of hyperthreading, as the 
super scalar architecture of Intel and AMD chips provides multiple 
instruction paths for integer operations only (aimed at gamers, they seem 
to outnumber us).  NONMEM is (almost) entirely floating point.  Dual core 
will multi thread very nicely (even for floating point operations), but 
that doesn't help a single NONMEM run.  But, as Darin points out, you do 
essentially get two processors, so you can run two models without loss of 
performance.  For those interested, a parallelized version of NONMEM may 
be available at some point, which would run twice as fast on a dual core 
(4 times as fast on a dual processor/dual core and potentially 100 times 
as fast in a data center with many processors)  See abstract from AAPS (
http://abstracts.aapspharmaceutica.com/ExpoAAPS05/CC/forms/attendee/index.aspx?content=sessionInfo&sessionId=1715)

Mark Sale M.D.
Global Director, Research Modeling and Simulation
GlaxoSmithKline
919-483-1808
Mobile 
919-522-6668
_______________________________________________________