Author: Vincent Diepeveen
Date: 18:11:56 05/14/04
Go up one level in this thread
On May 14, 2004 at 08:22:44, Robert Hyatt wrote: >On May 14, 2004 at 00:11:48, Vincent Diepeveen wrote: > >>On May 13, 2004 at 12:48:32, Robert Hyatt wrote: >> >>>On May 13, 2004 at 08:03:17, Vincent Diepeveen wrote: >>> >>>>On May 13, 2004 at 07:56:28, Vincent Diepeveen wrote: >>>> >>>>>On May 12, 2004 at 22:56:45, margolies,marc wrote: >>>>> >>>>>400 staff members, 12 megawatt, and just slightly faster than earth, >>>>>it all sounds stupid to me. >>>> >>>>400 staff members is of course too pathetic for words. 390 will be drinking >>>>coffee. >>>> >>>>>by the way, it won't be the fastest supercomputer at all. >>>>> >>>>>Note that they probably didn't order what they can get for their money in 2007 >>>>>either. clever salesmen promise something that's far under what they can deliver >>>>>in 2007 for that price. >>>>> >>>>>Suppose you would order for 50 million the fastest opteron cluster available by >>>>>then. >>>>> >>>>>Let's assume $6000 a node using myrinet pci-x. >>>>> >>>>>each node dual chip opteron. >>>>> >>>>>by then of course 1 chip has 4 cores. >>>> >>>>1 node has 4 cores. >>> >>>Or 1 node will have N chips, each of which will probably have 2-4 cores >>>depending on whether AMD delivers a 4-core opteron by then. >> >>I count at 2 a cpu. >> >> > > > >Count at whatever you want. dual core is 2005. 2007 will almost certainly be >quad-core... dual core opteron is *planned* for end 2005 :) I do not doubt it that they will make this happen, but for 2007 there is no plans for quad core opterons yet AFAIK and knowing developments must start years ahead and 2 is not comparable with 4, i foresee problems to make 2007 with it. Nevertheless i would be positively surprised if they do. If intel released in 2006 or 2007 a quad core P3, then of course AMD has to do it too. Please exclude itanium from the list. Note that last so many years the AMD and intel guys are no better than the average software guys. They didn't keep any schedule... > >>>> >>>>>4 processors * (50000k$/6k$) = 8333 nodes * 4 processors = 33333 processors. >>>>> >>>>>Right now 2.2Ghz single processor opteorn delivers 4 gflops. By then we have of >>>>>course 0.065 at 4Ghz or so and delivering more a cycle. >>>>> >>>>>So 10 gflop is not unrealistic. 33333 * 10 gflop = 333333 gflop = 333 tflop. >>>>> >>>>>So just ordering 50 tflop for 50 million now they get ripped off. What they can >>>>>deliver now for 50 million is of course not even close what they can deliver by >>>>>then for 50 million. >>>>> >>>>>By the way IBM already is building a 1 pflop machine. >>>>> >>>>>>OAK RIDGE, Tenn., May 12 (UPI) -- Energy Secretary Spencer Abraham said >>>>>>Wednesday the Oak Ridge National Laboratory will build the world's most powerful >>>>>>supercomputer by 2007 >>>>>>...... >>>>>>The supercomputer will be housed in a new 170,000 square foot facility that >>>>>>includes room for 400 staff members in 40,000 square feet of space. The machines >>>>>>will run on 12 megawatts of power, which will be supplied by the Tennessee >>>>>>Valley Authority.
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.