Author: Robert Henry Durrett
Date: 06:19:52 05/31/02
Go up one level in this thread
On May 31, 2002 at 06:57:18, José de Jesús García Ruvalcaba wrote: >On May 30, 2002 at 16:38:35, Robert Henry Durrett wrote: > >>On May 30, 2002 at 13:03:01, José de Jesús García Ruvalcaba wrote: >> >>>About what you propose: >>>1. It would work. >> >>Encouraging. >> >>>2. It would be somewhat inefficient. >> >>Yes. >> >>>3. Something better is going to play against Kramnik. >> >>Now you've got me interested! What do you suppose they might use against >>Kramnik? Everybody would like to hear this answer! >> > >An 8-processor machine with shared memory. That allows the eigth processors to >share the transposition tables without wsating too much time and reduces >significantly the amount of duplicated work (when compared to your proposal). Yes. All that appears quite reasonable. If their processors can be fast [i.e. 2.2 GHz clock rate at least], then that would surely be the best choice. I don't recall the specific bulletin, but another poster here posted a bulletin recently regarding the limitations of currently available 4-way and 8-way multiprocessor computers. They are slow!!! The principal limitation is that there are no currently available motherboards which can accommodate eight of the faster 2.35 GHz Pentium or 2.2 Xeon Intel multiprocessors. The same applies to the AMD processors. [Other processors are much slower.] Currently available 4-way and 8-way multiprocessor computers are REALLY slow. That is a problem! Chessbase is not going to want to use a slow computer for the Kramnik match if they can avoid it. If a currently available SLOW multiprocessor computer is used, the reduced speed might completely offset the advantages you expect from your more efficient design. The result might be that your 8-processor chess computer might actually prove inferior, if compared to the suggested ineffecient design [which can use the fastest multiprocessors], when measured in terms of it's expected performance rating. Do you have any information regarding the nature of the specific multiprocessors they will use? I would love to hear about it if you do. Will they be 32-bit? 64-bit? The 64-bit Itanium, 3rd generation [called Madison] should be available next Spring, and motherboards for 8-processor computers using these processors should be available then too. Programs like Crafty should be EXTREMELY strong on such computers. Unfortunately, these Madison processors are not available today, as far as I am aware. They are to use the new 9 mm technology and that's not commercialized yet. Maybe Intel is providing an advanced version of this to Fritz just for this match? Seems extremely unlikely. Too expensive. On the other hand, Intel would surely love to have the World Chess Champion beaten by an Intel-based chess computer. There would have to be a 64-bit version of Deep Fritz to use the Madison processors. > >>Note: An 8-processor computer is not going to be very impressive if the >>processors are slow. The whole idea of the configuration I was asking about was >>to get around the slowness of the processors. Each of the eight "computers" in >>the proposed "COMPUTER" could be VERY fast. > >And it takes a huge amount of work from the programmers. I do not doubt that >Frans Morsch can do it, but I have not heard he is working on that. >José. > >> Indeed, they could each be the >>fastest dual Xeon computers available. >> >>Bob D. >> >> >>>José.
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.