Author: Eugene Nalimov
Date: 16:33:51 01/22/00
Go up one level in this thread
Let's assume you have a fastest race car in a world, for example a jet car that can drive with a supersonic speed. Then you are replacing its engine by a VW-beetle (old one, or new one - doesn't matter) engine. Resulting car will be absolutely incompetive not only with "proper" version of itself, but with any "normal" car, including VF-beetle itself. To make it competive you'll have to completely redesign it - remove the wing that pushes it to the ground, because it is unnecessary on slow speed, redesign fuel system, wheels, etc. Without such extensive redesign car will never be competive with "normal" cars; more, even after it you cannot guarantee it will be the fastest one in its new class. And of course you'll not be able to make any predictions on "original" car behavior based on the behavior of the redesigned one. Designer of the original car may be not the best person for such a redesign, even if he is absolute genius in his target area. For example, he lacks the experience the designers of "normal" cars have (e.g. I doubt Seymor Cray will make more succesfull Aplle II design than Jobs and Wozniak). Nor let's return back to the topic: having search 1,000 slower will absolutely kill a lot of assumptions on which DB is built. They'll have to redesign it - i.e. introduce null moves (or something else based on a null move observation), get rid of singular extensions, then add some new specialized extensions that earlier were unnecessary because singular extension worked, remove some terms from the evaluation function to speed it up, maybe add some more speculative terms to compensate the shallower search depth, etc. It would be a tremendous amount of work. To do it they have to be strongly motivated - and they were already on top. Eugene On January 22, 2000 at 18:59:03, Tom Kerrigan wrote: >On January 22, 2000 at 17:55:28, Robert Hyatt wrote: > >>Please send a check for $500.00 to: >>I will send you one copy of a program that will run on your PC, and will >>say "hello from Deep Blue for the PeeCee" when you start it up. >>The point being that "Deep Blue for a PC would _not_ be Deep Blue, and >>I don't think _anybody_ would be fooled by the idea that it is... > >This is stupid and you know it. It is 100% possible to make Deep Blue for >Windows 98, it would just be a few thousand times slower than the "official" DB. >OF COURSE you will not confuse anybody into thinking that their $800 Gateway >Family PC is as good as IBM's Supercomputer, but that is not the point at all. >The point is that DB beat Kasparov, so it's incredibly cool to run the DB >algorithms on your PC. If FHH sold a PC version of DB, he will have thousands of >customers who aren't even sure how to play chess, but still think DB is great. > >>According to reported numbers, "DB" has 100K lines of C. Which would need >>serious porting since there is no SP-type hardware to use. And then it would > >So there's a lot of SP-ish assembly in it, or what? I don't see what the big >deal is. > >>need an eval written that takes about 40,000 clocks for the long path thru it. > >Actually, FHH should already have a fast version of the evaluation function in >software. Otherwise, how did he design it? How did he tune it? Presumably he >would want his thing to play as well/fast as possible vs. GM Benjamin... > >-Tom
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.