Author: Robert Hyatt
Date: 12:11:12 09/30/02
Go up one level in this thread
On September 30, 2002 at 04:25:47, Vincent Diepeveen wrote: >On September 29, 2002 at 23:48:39, Robert Hyatt wrote: > >You're on cocain Hyatt. I simply defer to my architecture books. I'm not on "anything" at all other than "facts"... > >A supercomputer by definition is something huge and big that's >having a tremendeous i/o speed. Not something where just 1 processor >which is a bit faster. Correct. _useful_ I/O speed. Very few consider having 1000 processors do independent I/O to 1000 different disk drives as really "useful" as a general-purpose computing tool. As compared to the Crays which don't resort to such inflated claims... > >I do not see crays to be very special. If i optimize DIEP for >a dual K7 then it will be taking only say a factor 10 or so to >make up for a processor or 32 crays, which never has many processors. 32 is not bad. A cluster of 8 crays with 32 cpus each is not bad... Of course you have to work to use them... > >Try to compensate for a 512 processor McKinley/Itanium2 SGI machine >with half a Terabyte hashtables. Nothing matches *that* speed >if you optimize for it. 512 GIGAHERZ. 512 gigahertz divided by the memory latency suddenly doesn't look that good. While a single cray CPU can read 32 bytes and write 16 bytes _every_ clock cycle... There is little to compare, in real-world applications... > >Of course you need to optimize for it. For the Cray you need to optimize >even more for to get working there. it is clocked at the same >Ghz like that McKinley is. 1Ghz. > >By *definition* a supercomputer is a big computer with a big i/o speed >and a bunch of processors. no it isn't, but I'll stick with the literature and let you coin your own definition as you choose... > >If you want to define 32 processor Cray as a supercomputer, that's obviously >fine. If you don't want to call a NUMA 512 processor SGI a supercomputer >then you're complete crazy. Actually, most call it a NUMA cluster. A big one, but a cluster. Which is a _far_ different animal from a true supercomputer... > >Best regards, >Vincent
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.