[Halld-pid] Fwd: New farm14 compute nodes arriving

[Mark M. Ito]

-------- Forwarded Message --------
Subject: 	New farm14 compute nodes arriving
Date: 	Thu, 6 Nov 2014 18:58:58 -0500 (EST)
From: 	Sandy Philpott <philpott at jlab.org>
To: 	Graham Heyes <heyes at jlab.org>, Jens-Ole Hansen <ole at jlab.org>, 
Robert Michaels <rom at jlab.org>, Harut Avakian <avakian at jlab.org>, Brad 
Sawatzky <brads at jlab.org>, Stephen Wood <saw at jlab.org>, Mark M. Ito 
<marki at jlab.org>, David Lawrence <davidl at jlab.org>, Veronique Ziegler 
<ziegler at jlab.org>, Maurizio Ungaro <ungaro at jlab.org>
CC: 	Patrizia Rossi <rossi at jlab.org>, Javier Gomez <gomez at jlab.org>



All,

The 104 new Haswell farm14 Physics nodes are in shipment, to arrive 
soon!  Two empty racks are being located in the F112 Data Center, next 
to qcd10g05 where they will be installed soon.  This is an addition of 
2496 cores to the current 1400 core farm.

Let the fun begin!
Sandy

------------------------------------------------------------------------
*From: *"Sandy Philpott" <philpott at jlab.org>
*To: *"David Rackley" <rackley at jlab.org>, "Kurt Strosahl" 
<strosahl at jlab.org>
*Sent: *Wednesday, August 27, 2014 8:36:49 AM
*Subject: *New farm14 nodes

*From: *"Chip Watson" <watson at jlab.org>
*To: *"Heyes Graham" <heyes at jlab.org>, "Sandy Philpott" 
<sandy.philpott at jlab.org>
*Sent: *Wednesday, August 27, 2014 8:34:10 AM
*Subject: *Re: New nodes

Graham,

The new nodes (88 bought by Physics, 16 bought by IT, total 104) are 
dual Intel E5-2670v3 (Haswell) 12 core, 2.3 GHz.  Each node has 32 GB 
DDR4-2133 memory.

The IT portion is intended to compensate Physics and LQCD for expected 
outages during the coming year as power and cooling are upgraded.

GlueX testing indicates that 2.3 GHz Haswell cores are 50% faster than 
the Sandy Bridge / Ivy Bridge cores already in the farm (as measured by 
events/sec/core on an 18 core part also at 2.3 GHz, with more memory per 
core).  That system scaled linearly through all 18 cores, so there will 
be no memory bandwidth bottlenecks at a mere 12 cores per CPU.

This will be the leanest memory configuration in the farm (less than 1 
GB per job slot on serial jobs), but we will try to run this with as 
many slots as experience shows it can handle (based upon statistical 
measurements in actual running).

Delivery expected late October, so production running by Thanksgiving at 
the latest.  More than half of the increase in capacity will be loaned 
to LQCD to build up a positive balance that can be drawn down in the 
second half of the fiscal year.

Chip


On 8/27/14 8:06 AM, Heyes Graham wrote:

    Hi Chip,
    	just to be sure that I have the numbers right could you email the specifications of the new nodes? Specifically how many processors per node, the number of cores per processor and processor clock.
    	Thanks,
    		Graham







-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/halld-pid/attachments/20141107/196e702d/attachment-0002.html>