AMD press release

Bob Drzyzgula bob at drzyzgula.org
Wed Nov 20 10:04:53 EST 2002


Perhaps you saw this announcement:

  http://www-916.ibm.com/press/prnews.nsf/jan/1D4E7167B5FB8C0E85256C7200484666

or some news about it, e.g.:

  http://www.internetnews.com/ent-news/article.php/1501201

or

  http://www.infoworld.com/articles/hn/xml/02/11/15/021115hnibmsuper.xml

Those are indeed very expensive boxes, almost as expensive
as the Windows boxes that Cornell uses :-) I'm not certain
who will buy them, but a perusal of the Top500 list, looking
for Sun- or HP-based clusters would probably give some hints.

As for the AMD processors, I don't know for certain, but
I'd be very surprised if it worked out this way. I expect
that AMD will be selling them as an "Itanium performance at
Xeon prices" kind of thing. My best guess is that Athlon
64 (Clawhammer) will be priced in the (small cache) Xeon
range, while Opteron (Sledgehammer) will be priced in the
(large cache) Xeon MP range, possibly lower, with the
32-bit Athlon left to duke it out with the Celerons and
non-Xeon Pentium 4s.  AMD is expected to discontinue,
or at least suspend, the Duron brand. Opteron systems
with four or more processors are in fact likely to be
fairly pricey, but it isn't clear that these will be what
you'd want to use for a Beowulf cluster anyway. Single-
and dual-processor systems are probably more appropriate
for most applications and are likely to be fairly reasonably
priced. At least, I hope.

My original question had to do more with the scientific
software, and the extent to which work has been done to
optimize the code for the x86-64 instruction set, along
the lines of what has been done for example in Intel's
Math Kernel Library, or the Sun Performance Library. The
response that my post has gotten seems to indicate that
this is a relatively small concern -- the out-of-order
execution capability of the processor, together with tools
such as GCC and ATLAS, are likely to give quite good
results right away.

--Bob

On Tue, Nov 19, 2002 at 02:11:43PM +0100, Alan Scheinine wrote:
> 
>    I lost the reference but someone at work indicated to me an
> URL which talked about a cluster of 128 Power4 for 78,000 dollars.
> I looked at the article and saw in the footnotes that 78,000 dollars
> was for four CPUs (plus memory, disk).  The article also said that
> the price of this cluster was good because it is similar to the
> price of Itanium 2 computers, about 78,000 dollars for a 4-processor unit.
> I find these numbers difficult to believe but my point is more general
> and more abstract.  Computers tagged as "servers" often have high prices
> when sold as complete computers.  This may be relevant the rumors that
> AMD wants to focus on the "server" market for the Opteron.  Does this
> mean there will be very few Opteron chips, available only as costly
> servers?  Bob Drzyzgula asked: What is the sense of list members as to
> when x86-64 will be a viable platform for scientific computing?  My own
> opinion is that the answer is "Now".  There is already scientific
> computing for the Alpha.  So my doubts regard the hardware aspect.
> For AMD 32-bit, Pentium and Xeon, I can get the motherboards made by
> Tyan and most of the motherboards on the Supermicro WWW pages.  But on
> those pages I do not see any Itanium 2 mother boards.  For low-cost
> clusters, Opteron motherboards sold like Pentium mother boards would
> be great.  But I have nightmares that if AMD can only produce a small
> quantity of Opterons, then they will seek to create a market in which
> an Opteron computer will be a source of very high profits.
> Alan
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list