SCIENTIFIC-LINUX-USERS Archives

September 2007

SCIENTIFIC-LINUX-USERS@LISTSERV.FNAL.GOV

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Troy Dawson <[log in to unmask]>
Reply To:
Troy Dawson <[log in to unmask]>
Date:
Wed, 26 Sep 2007 09:32:39 -0500
Content-Type:
text/plain
Parts/Attachments:
text/plain (64 lines)
Christoph P. Kukulies wrote:
> On Tue, Sep 18, 2007 at 09:30:37AM -0500, Troy Dawson wrote:
>> Christoph P. Kukulies wrote:
>>> We are considering buying a server for a memory intensive application.
>>> It should have at least 32 GB RAM and a Intel or AMD CPU.
>>>
>>> The application is single threaded, single CPU, that is, in no way
>>> optimized for multiple CPUS or multithreading, which does not mean it
>>> could be optimized for such in the future, but not at the moment.
>>>
>>> Memory throughput should be optimal. The data to be processed should
>>> reside all in RAM, paging should be avoided by all means.
>>>
>>> Would the following choice of hardware be supported by a 64-bit
>>> Scientific Linux?:
> 
> I found this article regarding memory throughput and a discussion of
> AMD-Opteron NUMA architecture. The conclusion is that only in 
> specially customized applications one can take advatange of the
> higher memory throughput.
> Otherwise the memory bandwidth is limited to the FSB BW, 6.4 GB/s.
> 
> The article can be found here:
> http://www.digit-life.com/articles2/cpu/rmma-numa.html
> 

The article is over two years old (the kernel has had additions since then to 
help both AMD and Intel processors), and while I didn't read it complete word 
for word, it never actually gave any real life tests, only their special memory 
throughput tests.
It really comes down to your application.  If possible, try it on two 
comparable CPU setups, one AMD and one Intel.  I've seen some wildly lopsided 
tests, try to at least give them the same amount of memory and the same disks.
Then run your application on it, and see which is faster.

That's how I decided I like the Opteron.  On my tests (recompiling rpm's) the 
Opteron beat the Xeon.  But I saw other people with the exact same setup, and 
for them the Xeon beat the Opteron.  It all came down to the application.

> 
>>>  o Tyan Tempest i5000PT (S5283) mainboard
>>>  o Intel XEON 5080 3730 MHz Diual Core CPU
>>>  o 32 GB RAM
>>>  o 3Ware 9590SE-8MI PCIe 4x
>>>
>>> Any comments welcome.
> 
> Just a last question, before I'm gonna place the order for the hardware:
> with the above Intel XEON 5080 CPU I would be able to address e.g. 32GB
> linearly and there is no hitch?
> 

I'm not the person to answer that.  I'm lucky if I ever get more than 2 Gig on 
a system.
In theory, if you are using x86_64 (not i386) yes, it should work without a 
hitch.  In reality ... I'll let someone who's tried let you know.

Troy
-- 
__________________________________________________
Troy Dawson  [log in to unmask]  (630)840-6468
Fermilab  ComputingDivision/LCSI/CSI DSS Group
__________________________________________________

ATOM RSS1 RSS2