 |
 |
X-MIMEOLE: Produced By Microsoft Exchange V6.5 Received: from exch-anon.win.slac.stanford.edu ([134.79.19.92]) by exch-mail4.win.slac.stanford.edu with Microsoft SMTPSVC(6.0.3790.1830); Tue, 31 Jul 2007 03:45:05 -0700 MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable Received: from nospam3.slac.stanford.edu ([134.79.18.83]) by exch-anon.win.slac.stanford.edu with Microsoft SMTPSVC(6.0.3790.1830); Tue, 31 Jul 2007 03:45:02 -0700 Received: from cernmxlb.cern.ch (cernmx08.cern.ch [137.138.166.172]) by nospam3.slac.stanford.edu (8.13.8/8.13.8) with ESMTP id l6VAj0Fu027755 for <[log in to unmask]>; Tue, 31 Jul 2007 03:45:01 -0700 (PDT) (envelope-from [log in to unmask]) Received: from cernfe01.cern.ch ([137.138.28.246]) by cernmxlb.cern.ch with Microsoft SMTPSVC(6.0.3790.1830); Tue, 31 Jul 2007 12:44:59 +0200 Received: from [128.141.129.201] ([128.141.129.201]) by cernfe01.cern.ch over TLS secured channel with Microsoft SMTPSVC(6.0.3790.1830); Tue, 31 Jul 2007 12:44:58 +0200 Content-class: urn:content-classes:message Subject: Re: The 8-core machines Date: Tue, 31 Jul 2007 03:44:57 -0700 Message-ID: <[log in to unmask]> In-Reply-To: <[log in to unmask]> X-MS-Has-Attach: X-MS-TNEF-Correlator: Thread-Topic: The 8-core machines Thread-Index: AcfTX9tDRIbx8UD3QqWjTFdMUZQk/A== References: <[log in to unmask]> <[log in to unmask]> From: "Fred Wickens" <[log in to unmask]> To: "Su, Dong" <[log in to unmask]> Cc: "Fred Wickens" <[log in to unmask]> Hi Su Dong, So far our experience with the 8-core machines has been good, =20 although we do not yet have a complete set of timings for each of the =20 algorithms. I confirm that the boxes we have installed at Point-1 are Dell 1950's =20 each with 2 x 1.86 GHz Clovertowns. I also believe that the next acquisition for the CERN Tier-0 will be =20 based on Clovertown chips, and given the way prices have developed I =20 assume they may well go to a higher clock speed. (Note that although they are made with clock speeds higher than 2.33 =20 the power consumption and price rise steeply above this, so I think =20 that for many large centres 2.33 GHz may be the upper speed =20 considered. Just for the record there is one at ~2.6GHz available to =20 all manufacturers and I believe Apple have sole rights for devices at =20 3.0 GHz for use in the big MacPro boxes, I don't know of anyone uses =20 either of these higher speeds in 1U servers.) One thing that is clear with the Clovertowns is that you really do =20 need to test your own application on them. Over the last few years =20 CERN IT had found that the relative performance of boxes with HEP =20 code (i.e. the mix of reconstruction and Monte Carlo) was reasonably =20 well measured by running SpecInt tests. This is no longer true with =20 the multi-core Intel processors and ratios vary a lot from one =20 application to another. For the AMD quad-cores we still have very little information. The =20 launch has been delayed and when they do appear (I think in =20 September) they will be a relatively slow clock speed. There are =20 hopes that they will ramp over to give something more competitive by =20 the end of the year. AMD has been very difficult to pin down on their =20 performance. However, I am quietly optimistic that they will be =20 competitive by early next year. Regards...Fred On 30 Jul 2007, at 17:42, Su, Dong wrote: > Hi Fred, > The SLAC tier-2 is now thinking about next acquisition of CPUs. =20 > There seems to be a leaning towards the 8 core machines similar to =20 > HLT got. We got the 2x4 1.8Ghz Intel Clovertown for HLT ? Any pros =20 > and cons learned from running them so far ? You are shooting for =20 > the AMD quad-core next ? > Regards, > Su Dong