p4 v itinium

Ricky Rankin r.rankin at qub.ac.uk
Sat May 18 04:04:36 PDT 2002


> You are comparing oranges to elephants.  It appears that Athlons will not 
> meet the needs of Mr Rankin due to the addressing limits that were mentioned. 
>  It sounds like the application will need large amounts of memory, fast 
> floating point, a very optimized fortran compiler and possibly high I/O 
> bandwidth which leads to Alpha Systems.  I do not find comparing the cost of 
> systems that will support the application vs the cost of unusable systems, 
> very helpful.  

Yes, it is a usable system that I am after. The budget is 
about£150K and the user base have problems, probably the 
majority, that will run well on p4 with 1-4GB memory. My own 
impression is that I need some combination of system that will 
cope with these small memory jobs and the larger memory jobs. 

Some vendors have been telling me using the p4 with Linux will 
enable the 4GB memory limit to be broken. If I read the 
responses correctly this appears to be sales speak and not quite 
factually accurate.

The p4 + a number of itaniums was of interest as from price 
performace angle. However again the correspondence on the list 
implies that the compilers and job scheduling software may not 
be as mature as my audience would expect. They would be looking 
for a stable and easy to use envirnoment.

Is the that fact that most of their codes require double 
precision floating point another indicator to move in the 
direction of the usual suspects of 64bit chips.

Thanks to the list for their input.


 > Best Regards,
> --George
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf

Ricky Rankin
Principal Analyst
Computing Services
Queen's University Belfast

tel: 02890 273819
fax: 02890 230592
email: r.rankin at qub.ac.uk

More information about the Beowulf mailing list