Author: Robert Hyatt
Date: 19:00:13 12/19/99
Go up one level in this thread
On December 19, 1999 at 20:00:17, Tom Kerrigan wrote: >Depending on your point of view, every chess program spends 100% of its time >doing alpha-beta searching. But if you break it down into move generation, >evaluation, etc., then alpha-beta is only a small fraction of that. So if you >can speed up everything else, you're really getting somewhere. > >The Deep Blue chips, despite all of the resources at hand, only worked with one >position at a time. If they were restricted to one position, an FPGA design will >definitely have the same restriction. > >The easy solution is to get a lot of FPGAs. =) > >-Tom > >On December 19, 1999 at 07:13:46, Dan Andersson wrote: >>I knew that, but the alphabeta search will dominate timewise. As you know fast >>movegeneration is exactly as important as fast primitive operations in a general >>purpose microprocessor. Fast is good but it doesnt mean algoritms will cange >>O(n). FYI I built gameplaying machines using RAM, EEPROM, ROM, TTY and whatnot >>using a soldering iron fifteen years ago. Kalaha and Reversi mainly. >> >>Regards Dan Andersson I always evaluate this a bit differently. IE if I had the idea to do my eval in an FPGA circuit (similar to 1978 Belle, for example) then the first question is what would I expect? If my eval was to be done in 1 pico-pico-second, I would run about 3x faster. That doesn't really attract my attention. By the time the FPGA is up and going, hardware will already be 3x faster, and my design will probably be too slow for the new hardware. In my code, eval is about 50% of total time, search is 10-20%. Which means if you leave out the "hard" part (doing alpha/beta in hardware) then I might go 5x faster. That is not bad, but that is reachable by jumping to the 21264 platform right now (for me). Doing the entire engine really offers the kind of speed advantage that makes this attractive.
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.