Author: Robert Hyatt
Date: 21:37:32 08/10/01
Go up one level in this thread
On August 10, 2001 at 16:28:11, Vincent Diepeveen wrote: >On August 09, 2001 at 22:09:29, Robert Hyatt wrote: > >>On August 09, 2001 at 20:41:30, Vincent Diepeveen wrote: >> >>>On August 08, 2001 at 12:50:07, Graham Laight wrote: >>> >>>>I don't think that xx86 --> Pentium range makes a good choice for computers with >>>>(very) large numbers of processors. IMO, large numbers of processors represents >>>>the best future option for increasing computer power in a cost-effective way. >>> >>>Have to disagree directly. No one can afford so many processors, >>>so much transport costs, so many risks, such a big power bill. >>> >>>Easier is a single processor being way faster and pressed cheaply >> >>The math doesn't work. For any single processor you build, I will take > >The math works perfectly for what i said. > >For intel or AMD it's easier to press 1 unit and sell it for 1000 dollar >than it is to press 2 units with a complex bridge in between and sell >it for 1000 dollar. > >That math will always be the same whatever statement you make. That math is nonsensical however. For _any_ chip you can "press" I can connect two for little more than the price of both chips. In fact, a dual MB costs less than two complete motherboards. So for _any_ processor you will ever be able to build, I can spend only 2X (or a bit less) and go twice as fast. You will _never_ be able to take a chip to 2x faster for only 2x the initial cost. It will be thousands of times that expensive, and you have to spread it over a bunch of chips to make it bearable. And the dual will always be twice as fast still. > >For a manufacturer making a faster processor is smarter than making >the current sucking chip parallel. That is simply nonsense to anyone doing silicon design. Why do you think they are doing 2-way superscalar? 3-way superscalar? > >Of course we as end users will keep on buying duals, that's a >different discussion which has nothing to with it here. > >The question whether in future we will see very cheap 16 processor >shared memory machines, that's very doubtful. It will happen. a 4-way machine is now very affordable. 10 years ago it would be horribly expensive. In 5 years there will probably be 4-way chips, rather than 4-way motherboards. > >Way more logical is a 16 times more powerful CPU! Ask any engineer about the cost of that 16 times more powerful CPU. And how many years it will take to produce it, compared to a 16-way machine based on today's readily available chips. No contest at all. Which has _always_ been why SMP machines are built. Cost-effectiveness, not anything else...
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.