Author: William H Rogers
Date: 10:32:21 02/12/01
Go up one level in this thread
On February 12, 2001 at 09:57:38, derrick gatewood wrote: >Recently I asked the question about distributive computer and chess programs and >got a healthy responce. I have fully investigated this matter and I am still >looking into it as I find it very exciting. I have a few questions. I have >done programming in the past, but I am mediocre. In the past few world >computer chess championships we have seen the entries of some very powerful >super-computers, such as CilkChess and P.ConNers. The silkchess was running on >a 256 processor computer with 32 gigabytes of shared memory, allowing very >extensive searches and large transposition tables. And P ConNers is also running >on a similar supercomputer. We have seen their amazing search results, such as >5-11 million nodes/sec! but more amazingly we see a shedder or fritz >consistently beat them. What is wrong with these supercomputer programs? Do >they lack the superior knowledge these other programs must contain? Or, even >without the knowledge programmed in, wouldnt the cilkchess eventually find the >better move with the use of its crazy search depth? How important is the nodes >per second--- wait, let me restate that, how important is the total nodes >searched? I used to have a pentium 350 and it would search only 2-300,000 nodes >a second, but it would beat some computers running the same program and getting >nps of about 500-700K. Once a certain narrow band of nodes are searched(the >most important nodes) are the hundreds of thousands or even million of nodes >searched after that only going to give a negligible result? >Is Crafty capable of running on a beowulf type cluster? or is the SMP version >mainly for boards that have quad/dual processor setups, like servers and such? >If the latter is true, then would it be easy to implement crafty in a beowulf >cluster with a minimal amount of programming? > >Thanks for all replies =) trust me, I am not wasting your time with these >questions.. I am trying to get somewhere, and I will only know once certain >questions are answered It could be that some of the programs that have very fast nps do not have that much chess knowledge. They operate on mostly tatical knowledge, (ie lots of captures) and lack positional knowledge. It was written by the earliest programmers that the more knowledge that you could put into a program, the better it would play, that is, if you had the most perfect knowledge, you could make the better move without the greatest look-a-head. Althoug many of todays top programs are closing the gap on perfect knowledge, most are not even close yet. I would be willing to bet that if one or more of the top programs could run a machine like Deep Blue did that they would probably beat the pants off of it. But then I am not a professional in the true sense of the word. Bill
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.