MLUG: [MLUG - DISCUSSION] VA Tech's "Big Mac", er "X" supercomputer *is* #3
[MLUG - DISCUSSION] VA Tech's "Big Mac", er "X" supercomputer *is* #3
Email address obfuscation in effect -- please click here to turn it off.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

So they now have official numbers from everybody. The Top500 press 
release is here:

    http://www.top500.org/lists/2003/11/press-release.php

While the list itself is here:

    http://www.top500.org/list/2003/11/

Interestingly, the VATech machine is now called just "X".  I suspect
either IP lawyers or a juicy promotional deal was at work here.
Mellanox is probably a very happy company today as well, since they
get equal billing on that site for their interconnect stuff.

In the "Fighting Words" department, The Apple-based cluster (2200
procs @ 2.00 GHz apiece) just eased ahead of the 2003 NCSA cluster,
which is based on 2500 3.06 GHz Xeons.  The G5 cluster is actually
almost 5% faster, but even if we call this a tie, then the result
does lend some merit to the suggestion that a 2 GHz G5 is definitely
in the same league as a 3 GHz Xeon, at least for clustered LINPACK
applications.

I'm certain that Intel will NOT be amused by this fact, so I think
it's safe to predict that we will see faster and better MP-capable
microprocessors from them in the near future, because they do NOT
want to be losing marketshare in this sector to the likes of IBM and
AMD.  There was only one Apple cluster in the Top500 this year, but
you *know* that's going to change, since their first effort was good
for #3 and was the cheapest of the Top10.  IBM is apparently on
track to produce a G5 chip for Apple that is at least 25% faster by
June.  Similarly, apparently only 4 of the Top500 systems were
Opteron-based this year, but, again, you know those numbers will go
way up next year.  At least 40 of the Top500 were built by SGI; I'm 
not sure what their trend is in HPC.

Oh yeah: obsolescence.  Here's the top-slot computer by year:

2003 #3  (VaTech, USA)
2002 #1  (Earth Simulator, Japan)
     #2  (ASCI Q, USA)
2001 #12 (Pittsburgh Supercomputing Center)
2000 #8  (ASCI White, USA)
1999 #27 (ASCI Red, USA)             # one of only 12 Top500 boxes
1998 #65 (ASCI Blue Mountain, USA)   # one of only 3 Top500 boxes
1997 #169 (T3E900 Cray, USA)
     #294 (T3E900 Cray, UK)   # these are the only 2 Top500 boxes
1996 and before: N/A

So from this data, it looks like you fall off the Top500 in 7 years
or less.  Usually much less; 95% of the computers on the list are
from 2001 or later.  In 2002, the Top500 list had 500 computers on
it, but *333* of them were bumped off or scrapped in 2003:

Date    # of surviving Top500 boxes

2003    333
2002    101
2001     26
2000     23
1999     12
1998      3
1997      2

Va Tech is planning their next move already...for 2006.  The odds
that they will have a Top10 computer next year are not that good,
now that we really know how to make *really* powerful cheap
clusters.  By 2005, they'll be well down the list, but by 2006 they 
project themselves having a machine about 5 times as fast as X, and 
(I'm guessing) 1100 machines to distribute to secretaries. :-)

jking


_______________________________________________
discussion mailing list
EMAIL:PROTECTED
http://mlug.missouri.edu/mailman/listinfo/discussion