Author: Vincent Diepeveen
Date: 04:49:10 08/04/05
Go up one level in this thread
On August 04, 2005 at 07:15:02, Engin Üstün wrote: >everybody understanding only like this: > >if i a am programming a simple engine and run it on a super hardware with 65535 >cores, this will be very strong play. Just prove you can get it to work at 8 cores first. >what is about the intelligence of the program? i mean the knowledge of the >program ? Well the competition in that respect has been closed already, as diep has more chessevaluation knowledge than any other program. That i'm a titled FIDE master didn't hurt of course when programming that knowledge. This is of course a gap you can never bridge. Too lazy even to make a parallel engine, let alone chessknowledge. Now people like to see how it plays all that chessknowledge, so do i. Right now i'm busy fixing chesspatterns i entered between 1994 and 1998. They are outdated simply and need to get improved, fixed and retuned to match the bugfree play that 2005 engines show. It's a pretty big job to improve all that old evaluation code (just kicking it out hurts even more by the way, as knowledge DOES work). >we human are not fast on thinking of position, but we are using our knowledge of >chess, and thinking selektive moves only very deep. >the goal is to go programming a better chess engine, and maybe more selektive >less nodes and with much chess knowledge. Diep searches at 85k nps at a k7 2.1ghz and it's move generator is worlds fastest move generator, so that nps is not because of bad programming at all. Yet i had also in 2001 a dual k7 (1.2Ghz) and have been crying out loud for many years that you first have to get that 12-14 ply before chessknowledge is more important to fix than another ply is important. Previous year at a quad opteron 2.0Ghz, diep searched worst case 10 ply search depth in several games. So it can't hurt coming few years to get a tad more plies with diep, to get a minimum of 14 ply. 4 cores isn't enough for the old versions. Of course i'm experimenting nonstop with forward pruning now in order to search deeper. If everyone has a big 8 processor opteron quad core, of course the chessknowledge and tuning is more important than whether you are good in making a dubious search that works for the hardware in question and is outdated the next year as there is a new cpu then. So if your only interest is in chessknowledge, then you should be in favour of parallel hardware. I wonder anyway what you are going to to at home to test your program when next year you have a quad core A64 at home. The A64X4. You still going to shout that everyone must run single core? >also is uniterested the next human vs mashine tournaments with super hardware >65535 core mashine. >that will be of course win the mashine.
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.