[Beowulf] Re: dual core Opteron performance - re suse 9.3

Joe Landman landman at scalableinformatics.com
Wed Jul 13 05:37:59 PDT 2005


Hi John:

   We find that we are talking about "per core" to our customers now.  I 
explain that previously, there has been an implicit 1-to-1 mapping 
between processor cores and chips, so that you could talk about either 
one and mean the other.  Now however, we are talking about per core, as 
things like licensing (lmgrd) aren't going to count chips or sockets, 
but will count cores.

   AMD uses the terminology of

	Np processors / Mc cores

so a dual Opteron 275 would look like

	2p/4c

system.  I prefer the converse of this, 4c/2p, but thats just me.  I 
don't know what (if any) terminology Intel uses for this.

Per socket is the same as per chip.  The issue is the terminology may 
not shift if you are talking about single core, dual core, quad core, 
... N core.  From an end user perspective, the cores are real full 
fledged CPUs that happen to share the same physical die as one or more 
other cores.  That is, with a little though, the end user can break the 
1-to-1 mapping and talk in terms of cores, in which case specifications 
start to make sense again.  1 GB per socket doesnt make much sense if 
each core needs 1 GB for a particular calculation.

Joe
John Hearns wrote:
> On Tue, 2005-07-12 at 21:00 -0700, Greg Lindahl wrote:
> 
>>On Tue, Jul 12, 2005 at 10:36:15AM -0500, Don Kinghorn wrote:
>>
>>
>>>The dual-core system had 4 one GB modules arranged 2 for each cpu.
>>
>>To be anal-hyphen-retentive, don't you mean "2 for each socket"?
>>
> 
> Acktcherly....
> we do need to decide on a terminology here.
> 
> I recently did a response to a tender for a prospective customer.
> I was tying myself in knots getting the correct terminology,
> for questions such as "the systems MUST have xxx gigabytes of RAM per
> processor"
> I went with reading that as 'per socket' in the case of dual cores.
> Also talking about 'dual nodes' is going to be more tricky.
> 
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf

-- 
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web  : http://www.scalableinformatics.com
phone: +1 734 786 8423
fax  : +1 734 786 8452
cell : +1 734 612 4615




More information about the Beowulf mailing list