Slashdot Mirror
25th TOP500 List Released
Chris Vaughan writes "The 25th edition of the TOP500 list of the world's fastest supercomputers was released today (June 22, 2005) at the 20th International Supercomputing Conference (ISC2005) in Heidelberg Germany. The No. 1 position was again claimed by the previously mentioned BlueGene/L System. At present, IBM and Hewlett-Packard sell the bulk of systems at all performance levels of the TOP500. The U.S is clearly the leading consumer of HPC systems with 294 of the 500 systems installed there (up from 267 six months ago)."
The list can be found here:6
http://www.top500.org/lists/plists.php?Y=2005&M=0
And here's a link to the actual list. Also interesting is the historical chart of the TOP500 by manufacturer, which tells a story in itself -- the decline of Cray and rise of IBM and Hitachi, for one.
You'd think that it would be a good idea to actually link to the html list, or the xml list, or the pretty charts.
The press release is interesting too.
Karma: SELECT `karma` FROM `users` WHERE `userid`=138474;
For you rabid fanbois (like me) here is how AMD scored:
/Processors Manufacturer Rmax Rpeak
Rank Site Country/Year Computer
10 Sandia National Laboratories
11 Oak Ridge National Laboratory
31 Shanghai Supercomputer Center
32 Los Alamos National Laboratory
33 Pittsburgh Supercomputing Center
39 US Army Research Laboratory (ARL)
46 Grid Technology Research Center, AIST
57 Swiss Scientific Computing Center (CSCS)
75 DOE/Bettis Atomic Power Laboratory
76 DOE/Knolls Atomic Power Laboratory
109 The University of Nottingham
144 Automotive Manufacturer (F)
155 Los Alamos National Laboratory
156 Government
167 Universitaet Wuppertal
174 United Institute of Informatics Problems
244 DaimlerChrysler
300 Veritas DGC
306 Ford Motor Company
347 Idaho National Engineering Laboratory
348 Japan Adv. Inst. of Science and Technology (JAIST)
388 Umea University / HPC2N
490 Chartered Semiconductor Manufacturing
499 Doshisha University
"The price good men pay for indifference to public affairs is to be ruled by evil men." ~Plato (427-347 BC)
It all depends on the system architecture and the type of problem being solved. Certain problems will adhere better to certain architectures and thus allow for a smaller gap between the theoretical and actual performance. The gaps can also be inherent in the architecture itself (e.g. communications bandwidth like you said).
These ranking are based on LINPACK doing traditional operations like solving linear equations, so supercomputers like the Cray MTA aren't even listed even though for some grand challenges they destroy everything else, for example when doing dynamic mesh weather simluations. Each processor on the memory grid has 128 processor threads where the active thread switches every cycle (so memory fetch has huge latency). This lets it have a unified memory model and still have extremely high throughput.
So the MTA can adjust the mesh to compute the tornado in very fine detail while using far fewer points for the huge swaths of calmer weather around it. Traditional supercomputers can't do that well since just distributing the data points to each processor is so much overhead.
Remember the goal of BlueGene is to build very dense systems. Not only do you have to factor in the costs of the system, but you have the costs of the facilities. This includes costs of construction or renovation of the facilities to handle the power and cooling requirements of these behemoths. BlueGene/L in it's current incarnation is using 32 cabinets for it's processors. While Earth Simulator is comprised of 320 cabinets for the CPUs (an additional 65 for interconnects).
More photos here. http://www.power.org/news/events/barcelona/photos/