Author: Slater Wold
Date: 22:35:06 02/19/04
Go up one level in this thread
On February 20, 2004 at 00:31:04, Keith Evans wrote: >On February 19, 2004 at 22:51:36, Slater Wold wrote: > >>On February 19, 2004 at 21:42:03, Keith Evans wrote: >> >>>On February 19, 2004 at 20:03:29, Bob Durrett wrote: >>> >>>> >>>>Microprocessors are lovable little creatures which are ADORED by all >>>>programmers, both male and female, because they are so easy [ : ) ] to program. >>>>[That, in fact, is part of the problem.] There is even a new breed of digital >>>>engineers who have wrapped their entire careers around the little cuties. Many >>>>programmers owe their very professional existence to microprocessors. >>>> >>>>It's all an evil deception intended to make programmers and engineers alike go >>>>astray. >>>> >>>>When microprocessors first became widely available, about thirty years ago, they >>>>hit the technical world like an atom bomb. People jumped on the microprocessor >>>>bandwagon like they were the best thing since sex and now some even worship >>>>microprocessors! The new programmers, scientists, and engineers just coming out >>>>of college think that microprocessors [and EPROMS] have been around forever, >>>>since before creation, and that it is a SIN to design anything which does not >>>>contain at least one microprocessor. >>>> >>>>It is the speed and sequential nature of microprocessors which is both their >>>>strength and their weakness, depending on the application. >>>> >>>>A chess programmer sees a microprocessor as being a gift from Heaven, along with >>>>the alpha/beta algorithm. [Shannon is seen as being a Saint.] >>>> >>>>If a chess engine were functionally decomposed into simple functional elements >>>>and if it were decided to provide hardware to perform those simple functions, >>>>then you can be sure that the modern designer would, without hesitation, reach >>>>for a microprocessor. Why? Because "That's the way things are done." Each >>>>functional element would have it's own dedicated microprocessor. >>>> >>>>Suppose the overall function of a chess engine were accomplished, mainly, by >>>>performing the various functions sequentially, one after the other. Suppose >>>>also that each function is performed by hardware elements each containing a >>>>microprocessor. What would happen? Since the functions would be performed one >>>>after the other [i.e. sequentially] and since each individual simple function >>>>would be performed by the sequential process within the microprocessor for that >>>>simple functional element, then the net result would be no faster or better than >>>>doing the entire chess engine function on a single microprocessor. To make this >>>>completely evident, note that I am postulating that only one microprocessor is >>>>working at any given time and that after one finishes the next starts. >>>> >>>>It should be evident that trying to create a hardware version of a chess engine >>>>should involve few if any microprocessors. Only those tasks which cannot >>>>possibly be performed non-sequentially should have a microprocessor. If more >>>>than one microprocessor must be used, then a way should be found for them to run >>>>in parallel. Better would be no microprocessors at all. >>>> >>>>The problem is that hardware designers skilled in digital design without the use >>>>of microprocessors is a breed of cat which may have long since become extinct. >>>> >>>>Satan laughs!!! >>>> >>>>Bob D. >>> >>>If you replaced all of the Xilinx FPGAs in Hydra with Opterons do you think that >>>it would get weaker or stronger? I vote for stronger. >> >>And which costs more? >> >>Xilinx, interestingly. > >It depends on which Xilinx part ;-) You ain't putting a decent chess program into a XC2V1000, we both know that. :) >I think that the XC2V1000 parts are around $200 now, but you can still fork out >say $7000 for an XC2V8000 part. I don't know how Chrilly gets his thing into an >XCV1000E part - I am curious about exactly what he has in there. The XCV1000E >has 1,000,000 marketing gates - if I remember correctly you typically have to >divide that by a factor of 3 to 8 or more to convert to realistic units. Xilinx >gurus can typically reduce that factor, but they have to be really intimate with >the Xilinx architecture and do a lot of hand tweaking. (In my book >http://www.andraka.com/ qualifies as a guy with "guru level" Xilinx knowledge.) > >-K
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.