Author: Ren Wu
Date: 09:43:39 08/09/99
Go up one level in this thread
On August 06, 1999 at 15:48:29, Paul J. Messmer wrote: >It actually plays correctly now, and for fun I've been playing it against Psion >chess (an old favorite) which it now is seems to be a little better than. That's cool. I also like Psion chess. I still remember that when i first saw it, on a PC/XT, it search 3 plies, and then start on 5 plies, i was really amazed, and somewhat shocked, how did he do that? >I'm >still mucking around with a lot of stuff, and the search and the static eval >(the eval consumes most of the time) is still written in C. On a single >processor PIII-600 I get around 400K nodes per second middlegame, in endgame >positions I can reach about 700K, which is nothing to write home about yet. Who >knows how this will change over time, though. There is much yet to be >optimized. Yes, the rate is about right, my program, initiative, doing about 320K on a PII 450. But to be honest, initiative is not optimized in a way that can maximize the use of L1 cache line. It is designed on a 486, and so can't be changed to take all advantage of the P5 ,and P6. I think that it is possible to get 600K on PII450 if i realy push it. Unfortunately, i was just too busy on other things. Also, i found now my program lost more games on bad positional judgement, so if i have more time, i may have to deal with the evaluation function first. >I had origially set out to make it as fast (NPS) as possible, and have the core >object code that takes 99% of the search time fit into the 16K L1 cache of a >PII/III. As I've worked with it more though, I wonder if it couldn't be much >stronger in the end with a larger "working set" than 16K and fewer NPS. to do this, you realy need push the speed, and stop adding knowledge. I am not sure too if this will weak your program, but you have to choose either go fo the speed or not. I was thinking to get rid of all second level knowledge, and play many games, and then *only* add those absolutely necessary second level knowledge. and see how far this can go. That's why i say i am really interested in your progress. I thought you are doing it this way. Have you use any MMX instruction? Any benefit from it? >I'm not a strong over-the-board or positional player, so I didn't want to try >and give it a ton of chess knowledge. I want something which isn't "smart", but >rather fast, deep, and perhaps selective. The AMD K7/Athlon processor I heard >has 64K code/64K data L1 cache, so maybe that processor will make my dilema a >moot point. Yes, but then you can't use your VTune! I was force to get a PPro in faver of a K6, just becaues of this. :) To push a processor to its limit, tools like Vtune is necessary. PS. I've looked your homepage, very interesting. I still have my smallest program with me (unfortunately, written by others, not me. :( ), for over 15 years, which has 8 bytes long! It actually can do a very good thing, to turn off the NumLock! It is really usuful back to old days that computer manufactures think that computer are only used by those people doing spredsheet, and not programmers. :) Ren
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.