Author: Robert Henry Durrett
Date: 16:37:13 06/18/02
Just went thru the SMP reference. Thx again to Dann for that. A few preliminary thoughts/perceptions: (1) People are interested in CM9000 because its programmers seek to provide a wealth of user benefits beyond raw power. There is a market for that. (2) It appears that getting the absolute highest SSDF performance rating would be achieved with SMP, rather than via Asymetrical Multiprocessing [AMP]. (3) But people are getting interested in, expecting, and demanding, MORE from their chess software. This trend should continue into a future where "Super CM9000" programs will be the rule rather than the exception. (4) Today, almost all the attention is on single processor machines, for various reasons. But the future may be different! (5) Multiprocessor machines will dominate if the price comes down enough and if the individual processors are fast enough. The multiprocessor machine must be significantly more powerful [or useful!!!] than single processor machines, or the added cost of multiprocessor machines will not be justifiable. Why asymmetrical? Asymmetrical multiprocessor computers would offer challenges not present in SMP, but they would offer much more programming flexibility. Which will be the most important in the future? I predict programming flexibility will be the preferred choice. About Programming Flexibility in AMP computers: Consider the analogous situation of a team of humans hired and organized to perform the computational tasks needed for any team objective/activity such as to play a chess game. It is standard practice to hire a team consisting of individuals with unique individual skills, knowledge, and other strengths such that the team members would complement each other nicely. The same strategy could be used in an AMP chess computer. Each processor, along with its associated software and hardware, could be optimized for certain essential tasks. The "essential tasks" assigned to one team member [i.e. processor with it's software/hardware] might be different from those assigned to another. In some cases, all team members might require certain essential skills. Some duplication of effort might be unavoidable, but it should be minimized, at least in most cases. Generally, one team member would be designated as the "task leader" or "task manager." The task manager would, in general terms, be responsible for the overall efficiency, performance, and output of the team. It would allocate task assignments and monitor task performance. As a simple example, one or more team members might be assigned primary responsibility for the user interface. It would be interesting to think about the way information might be transferred from one team member to another, and similar information-related issues. The issues associated with sharing a memory might fit into this category. I'm interested in seeing the ideas of others here at CCC who may be interested in thinking about the possibilities and difficulties that AMP might offer. Bob D.
This page took 0.01 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.