[Beowulf] itanium vs. x86-64

[Håkon Bugge]

There are several MPI2007 SPEC submission using Itanium and x86-64.  
Here is a comparison between two submissions using 128 MPI processes,  
comparing an SGI Altix 4700 using Itanium 2 Processors 9040 1.6GHz and  
the fastest x86-64 submission which use Intel Xeon E5472, 3.00 GHz, IB  
4x DDR, and Platform MPI 5.6.4. Both systems use 128 cores, whereas  
the Itanium system used 64 processors (chips) and the x86-64 system   
32 processors (dual vs. quad-core).

Obviously, this is not apples-to-apples, since we here compare a large  
SMP (that is a single node) vs. a cluster of 16 dual quad-core  
(Harpertown) nodes connected by ConnectX. The x86-64 system is 15%  
faster than the Itanium system based on the SPECmpiM(TM)_base2007  
metric (13.7 vs. 11.9). Here is a comparison on the individual  
benchmarks sorted on how many percent faster the x86-64 system is:


Benchmark	x86-64 % faster than ia64
107.leslie3d	-55.2%
137.lu	-25.8%
104.milc	-23.6%
130.socorro	-10.3%
127.wrf2	4.3%
128.GAPgeofem	13.9%
132.zeusmp2	15.5%
113.GemsFDTD	16.8%
115.fds4	23.6%
121.pop2	62.3%
126.lammps	73.7%
129.tera_tf	75.3%
122.tachyon	181.5%


The tendency here is clear, but not completely black-and-white. To me,  
this boils down to a YMMV discussion, essentially price performance  
for _my_ codes.

The complete data can be found here: http://www.spec.org/mpi2007/results/mpi2007.html



Thanks, Håkon


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.scyld.com/pipermail/beowulf/attachments/20090210/4d795c81/attachment.html