[Beowulf] "dual" Quad solution from Tyan

Karsten Petersen karsten.petersen at megware.com
Tue Feb 28 01:46:19 PST 2006


Hi,

just some remarks (I have a VX50 with 8x Opteron 880 in the other room...):

Mark Hahn schrieb:
>> http://www.cybernex.co.uk/tyan_vx50_quad_opteron_servers.htm
> afaikt, this is a ladder design: 
> 
> c-c-c-c
> | | | |
> c-c-c-c

Not quite, this would make no use of two HT channels.  Its like this:

c--c--c--c
|  |   \/|
|  |   /\|
c--c--c--c

(The two leftmost CPUs need their third HT to connect the two chipsets.)

> who says it's so great?  the HT fabric is pretty far from a "perfect"
> low-latency 16-port crossbar, for instance.  also, is it really cheaper?
> how about if you compare to four 2-socket DC nodes?

Well, from my experiences with the first sold VX50 systems:  If you have 
a well scaling MPI application you are probably better off in 
price/performance with a bunch of Opteron 180, 252 or 280 based machines 
and Infiniband.  (Or Intel Bensley...)  I consider these VX50 beasts 
more interesting for pre/postprocessing and not too large shmem/omp 
applications.  Such a machine is the answer to "16 threads - do I really 
need an Altix?"

Best wishes,
Karsten
-- 
HPC System Engineer

Tel.: +49 3722 528 43
Fax:  +49 3722 528 15
E-Mail: karsten.petersen at megware.com

MEGWARE Computer GmbH
Vertrieb und Service
Nordstraße 19
09247 Chemnitz/Röhrsdorf
Germany
http://www.megware.com





More information about the Beowulf mailing list