Comments on: Low Cost/Power HPC http://www.linux-mag.com/id/7799/ Open Source, Open Standards Sat, 05 Oct 2013 13:48:18 +0000 hourly 1 http://wordpress.org/?v=3.1 By: nir http://www.linux-mag.com/id/7799/#comment-8422 nir Wed, 30 Nov -0001 00:00:00 +0000 http://www.linux-mag.com/id/7799/#comment-8422 <p>The numbers in the comparison are mixed up in places -- the price ratio should be 22:1, the performance ratio 7.7:1, and the TDP 7.3:1 </p> The numbers in the comparison are mixed up in places — the price ratio should be 22:1, the performance ratio 7.7:1, and the TDP 7.3:1

]]> By: febbriaggne http://www.linux-mag.com/id/7799/#comment-8423 febbriaggne Wed, 30 Nov -0001 00:00:00 +0000 http://www.linux-mag.com/id/7799/#comment-8423 <p>quite surprising that the atom is more eficient tahn the Xeon.<br /> well, a quad core Xeon server is preferable than 22 mini ITX nodes i think.<br /> One of the problem is the cabling for connecting 22 nodes (if you build them independently), you will need a 24 port ethernet switch along with the cables to achieve the same performance with Xeon, instead of another problems like space requirement, and maintenance.<br /> Unless, someone start to think about multi processor atom server (a blade with 16 atom processors for example), which is absolutely much more neat. </p> quite surprising that the atom is more eficient tahn the Xeon.
well, a quad core Xeon server is preferable than 22 mini ITX nodes i think.
One of the problem is the cabling for connecting 22 nodes (if you build them independently), you will need a 24 port ethernet switch along with the cables to achieve the same performance with Xeon, instead of another problems like space requirement, and maintenance.
Unless, someone start to think about multi processor atom server (a blade with 16 atom processors for example), which is absolutely much more neat.

]]> By: deadline http://www.linux-mag.com/id/7799/#comment-8424 deadline Wed, 30 Nov -0001 00:00:00 +0000 http://www.linux-mag.com/id/7799/#comment-8424 <p>Thanks Nir!</p> <p>I got my numbers mixed up in my HTML table formatting. I fixed<br /> the table and now it is correct. I also re-checked all the numbers. (assuming I hit the right keys on my trusty old HP 15C). The big difference is that with POV the Xeon is 7.7 times faster, not 22 as I first reported. While this does not change some of the cost issues I mentioned, it certainly does put a handful Atom within striking distance of the Xeon. </p> <p>--Doug </p> Thanks Nir!

I got my numbers mixed up in my HTML table formatting. I fixed
the table and now it is correct. I also re-checked all the numbers. (assuming I hit the right keys on my trusty old HP 15C). The big difference is that with POV the Xeon is 7.7 times faster, not 22 as I first reported. While this does not change some of the cost issues I mentioned, it certainly does put a handful Atom within striking distance of the Xeon.

–Doug

]]> By: geoffwattles http://www.linux-mag.com/id/7799/#comment-8425 geoffwattles Wed, 30 Nov -0001 00:00:00 +0000 http://www.linux-mag.com/id/7799/#comment-8425 <p>Doug,</p> <p>Useful to anyone wishing to pursue this further is to examine the work over the decade. Significant <a>papers</a> by W. Feng at Los Alamos justify small clusters. Also, though not realized as yet, was the proposed Atom cluster <a>\"Molecule\"</a> at SGI.</p> <p>Thanks for covering this. I think you are on to important subject -- I remember the Cray 1 hiding it refrigeration 10x its size in the floor and benches surrounding it. </p> Doug,

Useful to anyone wishing to pursue this further is to examine the work over the decade. Significant papers by W. Feng at Los Alamos justify small clusters. Also, though not realized as yet, was the proposed Atom cluster \”Molecule\” at SGI.

Thanks for covering this. I think you are on to important subject — I remember the Cray 1 hiding it refrigeration 10x its size in the floor and benches surrounding it.

]]> By: linxmax http://www.linux-mag.com/id/7799/#comment-8426 linxmax Wed, 30 Nov -0001 00:00:00 +0000 http://www.linux-mag.com/id/7799/#comment-8426 <p>Well, I too was wondering whether the idea of disposing HPC’s would work with <a href="http://www.coop-systems.com/">bcm software</a> as you said in your previous post. The main advantage of it from what I see is that it can run really slow as it has its base on nodes which are pretty cheap by the way. Anyway, before we come to a conclusion that it would work as we expect, we need to take some factors into account. Since it is a low cost node, chances are that it would often get damaged and a need of constant replacement is going to be problem. </p> Well, I too was wondering whether the idea of disposing HPC’s would work with bcm software as you said in your previous post. The main advantage of it from what I see is that it can run really slow as it has its base on nodes which are pretty cheap by the way. Anyway, before we come to a conclusion that it would work as we expect, we need to take some factors into account. Since it is a low cost node, chances are that it would often get damaged and a need of constant replacement is going to be problem.

]]>