AMBER Archive (2003)

Subject: RE: Itanium II

From: Scott Brozell (sbrozell_at_scripps.edu)
Date: Wed Feb 12 2003 - 18:21:06 CST

Hello,

The Amber benchmarks page
http://amber.ch.ic.ac.uk/amber7.bench4.html
gives jac and gb_mb results for two Itanium2 machines.

On Wed, 12 Feb 2003, Ross Walker wrote:

> Dear Ramon,
>
> Here are some example timings I have for Amber 7. Unfortunately I only
> have single processor timings for an Itanium 2.
>
> Machine Specs are:
>
> Dual Processor 900MHz Itanium 2, 1GB Ram, 7200RPM IDE Disks, MSC-Linux
> IA64
> Single Processor (Hyper Threaded) 3.06GHz P4, 1GB Ram (533MHz), 7200 RPM
> IDE Disks, Redhat 8.0
>
> Job description - Sander v7 - MD LADH approx 76000 atoms, 12A cutoff,
> PME, 1fs Step, No Shake, 500ps.
>
> 1Processor IA64: Intel Compiler v7 with Itanium2 optimisations: = 1902.0
> seconds
> 1Processor IA64: G77 2.96 with standard optimisations = 8217.0 Seconds
>
> Note how poor the performance is with the GNU compilers, I suspect this
> is due to the system running some kind of 32 bit emulation - the intel
> compilers run in full 64bit mode.

For sander g77 version 2.96 yields significantly worse performance
than version 3.1.1 on many platforms and for many kinds of calculations.
For Itanium2 platforms, where software pipelining will be important
for performance, I suggest a workstation type compiler; under Linux that
would be efc. Free Non-Commercial Unsupported Intel compilers can be
downloaded from
http://developer.intel.com/software/products/compilers/flin/noncom.htm
 
Scott Brozell

> This seems reasonably quick, however if you compare it to a 3.06GHz P4,
> (which is about a tenth the price):
>
> 1Processor P4 (HT switched off): G77 v3.2 with standard optimisations =
> 2956.0 seconds
> 1Processor P4 (HT switched off): Intel Compiler v7 with P4 optimisations
> = 2474.0 seconds
>
> So, while the Itanium2 shows good performance for the clock speed it
> loses out to the much cheaper and higher clocked 32 bit P4 processors,
> Intel should have a 3.3GHz P4 out soon that will come very close to the
> Itanium time. Thus my opinion is that Itanium2's are too expensive when
> you consider the price performance metric. A dual P4 Xeon would probably
> offer better performance at a fraction of the cost.
>
> Hope this helps
> All the best
> Ross
>
>
> > -----Original Message-----
> > From: Jose Ramon Blas [mailto:jramon_at_ub222059.pcb.ub.es]
> > Sent: 11 February 2003 18:51
> > To: AMBER MAIL REFLECTOR
> > Cc: VMD list
> > Subject: Itanium II
> >
> >
> > Hi,
> >
> > does anybody has experience on the performance/speed of the
> > new Itanium II
> > processor on molecular dynamics simulations??
> >
> > Thanks for your comments,
> >
> > Jose Ramon