Author: Bas Hamstra
Date: 11:49:46 09/02/99
Go up one level in this thread
On September 02, 1999 at 12:41:31, Steve Maughan wrote: >Bas > >Interesting - I think I'll have tinker with some of your idea! I'm using Delphi >so I'll have to try and create an assembler function for the FirstSetBit. > >Just out of curiosity - how many nodes per second do you get with this >architecture / implementation? > >Regards > >Steve Maughan Delphi is perfectly ok. I use CBuilder 4 which is the same but uses C/C++ in stead of Pascal. I don't know about your Delphi, but this BCB allows inline assembly code. How many NPS? With hardly any eval (basically material with some piece square eval code to centralize pieces) it hits 200-250k NPS on a Celeron 466, with full SEE sorting. That's in the same ballpark as Crafty which does 100-150 kNPS on the same CPU. If I leave out the check evasions in the qsearch it goes quite a bit faster (30% or so). With a normal eval I expect it to have around the same speed as Crafty. Another version which uses bitboards is still slightly faster. However optimized to the core I think the versions come very close. Also the bitboard version has a stripped make/unmake (without function calls) so that explains some of the difference. I'm not saying its the fastest possible method, I just think it's competitive. Some people say you're crazy if you do incremental attacks. I don't think so at all. It has another interesting possibility: you can generate 1 capture at a time. Just take the most valuable victim and pop out the least valuable attacker on it. I am not sure the extra speed outweighs the bad sorting, especially at greater depths (no more SEE sorting, however you can still toss out SEE wise losers). However I stopped doing that, because it's too messy.
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.