[Beowulf] Register article on Opteron - disagree
Douglas Eadline, Cluster World Magazine
deadline at linux-mag.com
Mon Nov 22 14:09:29 PST 2004
My top500 rant.
Using the Top500 to draw any kind of market trends and conclusions is like
standing outside of a Hummer dealership and asking customers if gas
millage is important.
A little data from the other end of the HPC street may be helpful. On my
very unscientific polls at ClusterWorld.com, I asked recently "how big is
Of the 162 responses, 52% said their cluster was under 32 nodes.
Furthermore I also asked, what processor you expect your next cluster to
53% said Opteron.
How accurate is this data? I don't know, but if the Register is going to
draw lines through single data point, then I'll take the liberty and say
that according to my little survey, most clusters are less than 32 nodes
and therefore do not show up on the top500 list and are uncounted. In
addition, Opteron seems to be the processor of choice for clusters.
Honestly, I don't know how accurate these numbers are, the samples are
small, and have no way of knowing how random the sampling is.
I think the next survey question I'm going to ask is how well does my
application(s) scale. That is, what is the maximum number of processors
your code can use efficiently. Because, looking at benchmarks for
a number of processors an order of magnitude higher than your limit is
well, silly. Of course some applications can use that many processors,
but I am guessing that most don't. (You know that whole Amdahl's law thing)
On Sun, 21 Nov 2004 john.hearns at clustervision.com wrote:
> I spotted this article on The Register.
> As someone who has installed Opteron clusters, including what I believe was the
> first in the UK (used for computational chemistry, and going strong),
> I disagree with this article.
> Opteron are alive and kicking.
> What's more the article confuses 'supercomputer' with '(super)computer in
> the Top 500'
> and also makes no mention of EMT64/Nocona. The artcle says Xeon/Itanium,
> then goes on to talk about 32 bit Xeon.
> The fact that there are fewer Opteron based systems in the Top 500 is
> irrefutable (I didn't know this) but it makes me uneasy to extrapolate
> this to the impending death of a CPU.
> I DO agree (and let's have some debate here) that Nocona is bound to make
> big inroads.
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
Editor-in-chief ClusterWorld Magazine
Cell: 610.390.7765 Redefining High Performance Computing
Fax: 610.865.6618 www.clusterworld.com
More information about the Beowulf